- Email: [email protected]
Emission timing control method for improving signal to interference ratio on public address system
Applied Acoustics 98 (2015) 70–78
Contents lists available at ScienceDirect
Applied Acoustics journal homepage: www.elsevier.com/locate/apacoust
Emission timing control method for improving signal to interference ratio on public address system Taira Onoguchi ⇑, Dan Murakami, Yoshifumi Chisaki Graduate School of Science and Technology, Kumamoto University, 2-39-1 Kurokami, Chuo-ku, Kumamoto 860-8555, Japan
a r t i c l e
i n f o
Article history: Received 11 September 2014 Received in revised form 20 March 2015 Accepted 30 April 2015 Available online 18 May 2015 Keywords: Public address system Sound overlap Emission timing control Computer network Disaster mitigation
a b s t r a c t In order to reduce disaster damage, there are several media such as email, TV, radio and a public address system with outdoor loudspeakers to deliver emergency announcements to the people at risk. The Great East Japan Earthquake revealed the importance of a public address system with outdoor loudspeakers for emergency announcements and the difficulty of its intelligibility. To achieve a public address system with sufficient intelligibility for residents, one important factor is to suppress sound overlap caused by reflection from terrain or buildings, or caused by the simultaneous emission of other loudspeakers. Recent Internet technologies and devices will contribute to avoid sound overlap caused by the simultaneous emission of other loudspeakers. In this paper, a method of controlling emission timing to minimize the sound overlap caused by the simultaneous emission of other loudspeakers of a public address system is proposed. This method focused on the transmission of a signal between the input and output of a public address system. Under the assumptions that the attenuation and delay factor between a loudspeaker and a listening point depend only on the distance and the directivity of each node is omnidirection, the optimal emission timing for a pair of nodes is expressed in terms of the signal to interference ratio. Then, on the basis of the above optimal emission timing, the emission timing for a set of nodes in a management area is determined. In the proposed method, each node collects and shares the positions of each node obtained by the Global Positioning System via a computer network to determine the emission timing, and then taking into account whether other nodes emit sound or not each node autonomously determines its emission timing and the delay time for other nodes. The proposed method can be implemented without any dedicated hardware and can, therefore, be applied to existing loudspeakers. The delay and error factors in a public address system employing the proposed method are also modeled and estimated. Ó 2015 Elsevier Ltd. All rights reserved.
1. Introduction In order to reduce disaster damage, there are several media such as email, TV, radio [1,2] and a public address system with outdoor loudspeakers shown in Fig. 1 to deliver emergency announcements to the people at risk. In the Great East Japan Earthquake, among residents who could obtain evacuation announcements in the affected prefectures, 45% of them obtained it from outdoor loudspeakers. However, among those who obtained evacuation announcements from outdoor loudspeakers, 56% of them answered they could clearly hear the announcements [3]. Therefore, the Committee for Technical Investigation on Countermeasures for Earthquakes and Tsunamis Based on the Lessons Learned from the ‘‘2011 off the Pacific coast of Tohoku Earthquake’’ suggests improvement of ⇑ Corresponding author. Tel.: +81 96 3423621; fax: +81 96 3423630. E-mail addresses: [email protected] (T. Onoguchi), chisaki@cs. kumamoto-u.ac.jp (Y. Chisaki). http://dx.doi.org/10.1016/j.apacoust.2015.04.019 0003-682X/Ó 2015 Elsevier Ltd. All rights reserved.
intelligibility of a public address system as one of the countermeasures to mitigate tsunami damage [4]. To achieve addressing with sufficient intelligibility for residents, factors that degrade the intelligibility of radiated sound must be minimized. One important factor is sound overlap at observation points. In particular, sound overlap caused by reflection from terrain or buildings, or caused by the simultaneous emission of other loudspeakers, is known to degrade intelligibility [5–7]. As demonstrated by the hearing ability of humans, and illustrated by the cocktail party effect, intelligibility depends not only on the power ratio between the target signal and interference but also on the context of the speech, the characteristics of the signal, the direction of arrival, and word familiarity. However, in the previous work it was shown to be possible for intelligibility to be estimated from the power ratio of overlapping sounds [8]. In accordance with the result of this previous work, in this paper, sound overlap is defined by the power ratio of the signal arrived from one node to the signals arrived from other nodes. Therefore, in this paper, the impact of the context of the speech,
71
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78
This paper is organized as follows. In Section 2, the optimal emission timing for a pair of nodes in terms of the signal to interference ratio is derived. Then a model for determining the emission timing for a set of nodes in a management area is shown. In Section 3, to accomplish the optimal emission timing derived in Section 2, a method of controlling emission timing for a public address system is proposed. In Section 4, the delay and error factors in a public address system employing the proposed method are investigated. In Section 5, the proposed method is implemented on a laptop PC without any dedicated hardware and a demonstration is conducted. Conclusions are given in Section 6. 2. Model for determining emission timing for public address system 2.1. Signal to interference ratio in public address system
Fig. 1. Outdoor loudspeaker of a public address system for emergency use.
the characteristics of the signal, the direction of arrival, and word familiarity are not considered. Because the signals emitted by the nodes of a public address system in one area are the same, when the phase of overlapped sounds is the same, the sound is emphasized rather than masked. However, in the actual situation, this kind of phase synchronization can be considered as a rare case. Therefore, in this paper, the impact of phase is not considered. Based on this definition of sound overlap, sound arriving from adjacent loudspeakers can be considered to degrade intelligibility. Therefore, a method of suppressing sound overlap between adjacent loudspeakers is required. One simple way is to set loudspeakers so that each service area of them does not have overlap. However, since there are loudspeakers already installed, it costs so much to move them. For example, at the end of 2001, 65% of all the local governments of cities, towns, and villages in Japan was equipped with the broadcast system for disaster [9]. In this paper, a method for controlling the timing of emission to reduce sound overlap in a public address system is proposed. In a public address system with loudspeakers, an input signal is transmitted via a path consisting of two portions. The first portion is the path between the input and output of the public address system. The second portion is the path between a loudspeaker and a listening point. The latter path is affected by many factors such as rainfall, absorption by air and terrain, the distribution of temperature, humidity, reflection by terrain and buildings, and wind. In particular, previous works showed that open-air sound transmission between a loudspeaker in a public address system and a listening point are affected by reflection [6] and wind [10,11]. In this paper, the latter path is simplified such that attenuation and delay factors of the impulse response between a loudspeaker and a listening point only depend on the distance. The proposed method focuses on controlling the emission timing for the former path, the path between the input and output of the public address system. The proposed method delays start time of emission of each node to avoid sound overlap by following processes. Each node collects and shares the positions of each node obtained by the Global Positioning System (GPS) via a computer network to determine the emission timing. An input signal can be fed into one of each node in the same management area. The input signal is transmitted to other nodes, then each node detects the start time of a sound event from the transmitted signal. After detecting the start of a sound event, a node sends a query to other nodes to obtain their status of emission. On the basis of response from other nodes, each node autonomously determines its emission timing.
The amplitude of an input signal fed into a public address system at time t is denoted by
aðtÞ ¼
A; 0 6 t 6 N 0;
otherwise
;
ð1Þ
where A 2 R and N is the duration of the input signal. The amplitude of the output signal of node sn at t, denoted by bn ðtÞ, is obtained as
bn ðtÞ ¼ g n aðt sn Þ;
ð2Þ
where g n is the gain factor and sn is the delay factor of sn . Then the amplitude of an observed signal at a listening point in the service area of sn at time t, denoted by yn ðtÞ, is calculated as
yn ðtÞ ¼ ¼
1 X
bn ðkÞhn ðt kÞ;
ð3Þ
g n aðk sn Þhn ðt kÞ;
ð4Þ
k¼1 1 X k¼1
where hn ðtÞ is the impulse response between sn and the listening point. yn ðtÞ can be rewritten as
yn ðtÞ ¼ g n an aðt sn bn Þ;
ð5Þ
where an is the attenuation factor and bn is the delay factor of hn ðtÞ. an and bn depend on many factors such as distance attenuation, rainfall, wind, reflection by terrain and buildings, absorption by air and terrain, the distribution of temperature, and humidity. In this paper, an and bn are assumed to depend only on the distance as follows:
rn yn ðtÞ ¼ g n aðr n Þa t sn ; c
ð6Þ
where r n is the distance between sn and a listening point and c is the speed of sound. Therefore, the signal to interference ratio at time t regarding the sound from sn as the signal and the sound from other nodes sm ðm – nÞ as interference, denoted by SIRn ðtÞ, can be obtained as follows:
jy ðtÞj SIRn ðtÞ ¼ 20log10 P n ; m jym ðtÞj g aðr n Þjaðt sn rcn Þj ¼ 20log10 P n rm : m g m aðr m Þjaðt sm c Þj
ð7Þ ð8Þ
The vector of the optimal delay factor for each node in terms of the signal to interference ratio, sopt , is then represented as
XXX SIRn ðtÞ;
sopt ¼ argmax sn ;sm
n
ð9Þ
S t2tn
P where tn is the vector of time when yn ðtÞ – 0 and S is the summation over listening points in the service area of sn .
72
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78
2.2. Optimal emission timing for pair of nodes in terms of signal to interference ratio For a pair of nodes sa and sb with the delay factor satisfying sopt , the following condition is required:
rb a t sb ¼ 0; c
ra if a t sa – 0: c
ð10Þ
Because aðtÞ ¼ 0 for t > N according to Eq. (1), this requirement can be converted to the following equation:
sa þ
ra rb > sb þ þ N: c c
ð11Þ
Therefore,
sa > sb þ
rb ra þ N: c
ð12Þ
It is sufficient to satisfy Eq. (12) only at the listening point on the near-side edge of sb for sa ; Ladj . The position of Ladj depends on the pattern of overlapping service areas. Fig. 2 shows patterns of overlapping service areas. In this paper, the directivity of each node is assumed to be omnidirection for simplifying. Therefore, with the assumption that an depends only on the distance, the service area of each node is assumed to be a perfect circle. Hereinafter, the diameter of the service area of each node is denoted by R and the distance between sa and sb is denoted by dab .
2.3. Determination of emission timing for set of nodes in management area Node sn is represented by
sn ¼ ½ pn
cn ðtÞ en ðtÞ ln ðtÞ ;
ð14Þ
where pn is the value for the priority of sn . Priority is preliminarily given to each node taking factors such as population into account. Duplicate priority values are not allowed in the same management area. Also, a node located near the border of a management area must not have a priority value that is duplicated in each adjacent area. cn ðtÞ is the emission count of sn up to time t. en ðtÞ is a flag that takes a value of 1 when sn emits a sound at time t and 0 when sn does not emit a sound at t. ln ðtÞ is the most recent finish time of the emissions of sn . Hereinafter, Sn represents the service area of sn and Om represents an overlapping service area. If there are N nodes and M service areas overlapping in a management area, matrix B is defined by
2
b11 6 . 6 . 6 . 6 B¼6 6 bn1 6 . 6 . 4 . bN1
b1m .. .
bnm .. .
bNm
3 b1M .. 7 7 . 7 7 bnM 7 7; .. 7 7 . 5
bnm ¼
1; Sn Om 0;
otherwise
:
ð15Þ
bNM
The situation in Fig. 3, for example, can be represented as
(1) Pattern A (dab < R). qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 2 At Ladj , r a ¼ ðR dab Þ þ ðH hÞ and r b ¼ R2 þ ðH hÞ ,
ð16Þ
where H is the height of the loudspeaker and h is the height of the listening point. Therefore, Eq. (12) is converted to
sa > sb þ N
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 2 R2 þ ðH hÞ ðR dab Þ þ ðH hÞ
Matrix O is defined by
ð13Þ
2
(2) Pattern B (R 6 dab 6 2R). qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 2 At Ladj , r a ¼ ðdab RÞ þ ðH hÞ and rb ¼ R2 þ ðH hÞ .
6 6 6 O¼6 6 4
þ
c
:
Therefore, Eq. (12) is again converted to Eq. (13). (3) Pattern C (dab > 2R). There are no service areas overlapping. Therefore, no requirement is derived from Eq. (12).
Pattern A
Pattern B
Pattern C
sα
sβ
O1
0
0 .. . 0
O2
3 0 .. 7 . 7 7 7: 7 0 5
..
. 0
ð17Þ
OM
Then matrix U is defined as
2u 3 1
6 .. 7 6 . 7 6 7 6 7 U ¼ 6 un 7 ¼ BO: 6 7 6 .. 7 4 . 5
ð18Þ
uN
Ladj
: Service area Ladj
2
Ladj rα
4
1
rβ
H 2
h dαβ
3
1
R
Fig. 2. Patterns of overlapping service areas.
Fig. 3. Example of overlapping service areas.
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78
The flag indicating whether sn can emit sound at t; v n ðtÞ, can be obtained from matrix U and operator VðtÞ as
v n ðtÞ ¼ VðtÞun ¼
1; 8k 2 k; Gðt; Ok Þ ¼ 1; or k ¼ ; ; 0; otherwise
ð19Þ
where k is a vector consisting of the indexes of overlapping service areas included in un . If Ok ¼ Sn \ Sq , the operator G returns a value such that
8 1; eq ðtÞ – 1; and t lq ðtÞ > tnq ; and > < p pn þ cn ðtÞ < maxðpqn ;pq Þ þ cq ðtÞ ; Gðt; Ok Þ ¼ maxðpn ;pq Þ > : 0; otherwise
ð20Þ
where tnq is the required difference in emission timing between sn and sq to avoid the overlap of emitted sounds. The above conditional equation allows a node with a smaller priority value to emit sound earlier. According to Eq. (13), t nq is obtained as
t nq ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 2 R2 þ ðH hÞ ðR dnq Þ þ ðH hÞ c
ð21Þ
:
3. Method of controlling emission timing for public address system To accomplish emission timing control based on the emission timing determined in Section 2.3 in accordance with Eqs. (20) and (21), each node must obtain the following information. Duration of input signal. Distance between nodes. Priority, emission count, the flag indicating whether or not emission is being conducted now, and the most recent finish time of the emission of other nodes with which the service area overlaps. Precise time information. Because a signal transmitted over a public address system is initially unknown, each node must utilize some method to detect the start and end of the sound event to determine the duration of the input signal. To obtain the distance between nodes, each node must obtain information on its location. Also, the information listed above must be shared among the nodes in a management area. A method of controlling the emission timing for a public address system that meets these requirements is proposed in this section. A schematic diagram of the proposed method is shown in Fig. 4. A block diagram and a flowchart of the proposed method are shown in Figs. 5 and 6, respectively. The proposed method utilizes one server and N nodes, s1 ; s2 ; . . . ; sn ; . . . ; sN , for a management area. A server consists of a database containing information on the nodes in the same management area and network devices used to exchange information with nodes. Each node consists of loudspeakers, a microphone to input the signal, a receiver for warning Server of each management area
systems such as the national early warning system (J-ALERT) [12], a database containing information on other nodes, a buffer to store input signals, a GPS to obtain its location and precise time information, and network devices to exchange information with the server and other nodes. The priority value pn is preliminarily given to sn taking factors such as the local population into account. Duplicate priority values are not allowed in the same management area. A node located near the border of a management area must not have a priority value that is duplicated in each adjacent area. Under the assumptions that the attenuation and delay factor between a loudspeaker and a listening point depend only on the distance and the directivity of each node is omnidirection, the service area of each node is assumed to be a perfect circle with diameter R. 3.1. Selection of active network for information exchange Each node can select one active network from several types of available network, such as WiFi, 3rd generation (3G), Long Term Evolution (LTE), or a mesh network, for backup in the case of a network being down. The selected network is used for information exchange between the server and nodes and between nodes. 3.2. Collection of own information by node and construction of database of other nodes Node sn obtains information on its own location from a GPS or preliminary measurements. This information consists of the latitude /n , longitude kn , geoid height Hgn , and elevation En . These values are converted to earth-centered earth-fixed (ECEF) coordinate values xn ; yn , and zn as follows:
xn ¼ ðN þ En þ Hgn Þ cos /n cos kn ;
Network devices GPS
ð23Þ
zn ¼ ðNð1 e2 Þ þ En þ Hgn Þ sin /n ;
ð24Þ
where N is the radius of the prime vertical and e is the first eccentricity. After calculating the coordinate values, sn sends them and its priority value pn to the server in the same management area. If node sn is located near the border of another management area, it sends
Network devices
Receiver for warning systems Node
Network devices GPS
Mic.
Receiver for warning systems Node
Fig. 4. Schematic diagram of proposed method.
ð22Þ
yn ¼ ðN þ En þ Hgn Þ cos /n sin kn ;
WiFi, LTE, 3G, mesh network
Mic.
73
Fig. 5. Block diagram of proposed method.
74
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78
such as J-ALERT. Then the input signal is transmitted to other nodes and stored in the buffer of each node. Each node reads the signal from its buffer. If it detects the start of a sound event, it starts the following process to determine the emission timing.
Read input signals from buffer Start of sound event is detected ?
No
3.5. Determination and control of emission timing
Yes
Send query to other nodes which overlap service area
vn = 1 ?
No
Wait tp (s)
After detecting the start of a sound event, node sn sends a query to other nodes with an overlapping service area, sm ðm – nÞ, to obtain the current values of pm ; cm ; em , and lm described in Section 2.3. If sn does not obtain a response from a node within an arbitrary time limit, this node is considered as being down and its state is ignored in the determination of the emission timing. Depending on the response from sm , Eq. (19) is evaluated. In this process, owing to error factors, the following equation is used instead of Eq. (20):
n
Exceeds No
tlim ?
Yes
Yes
Emit signals Yes
End of sound event is detected ?
No
8 1; eq ðt c Þ – 1; and t c lq ðtc Þ þ t sync > tnq þ t GPS ; and > < p pn þ cn ðtc Þ < maxðpq ;p Þ þ cq ðt c Þ Gðtc ;Ok Þ ¼ ; maxðpn ;pq Þ n q > : 0; otherwise ð26Þ
Yes
Send end signal to other nodes Fig. 6. Flow chart of proposed method. v n is the flag of emission timing determination, tpn is the waiting time until resending of a query, and tlim is an arbitrary time limit for waiting.
the coordinate values and priority value to each server in the adjacent area. After receiving this information, the server registers the coordinate values, priority value, and IP address of sn with a time stamp to the database. Then, the server sends the coordinate values xm ; ym , and zm , the priority value pm , and the IP address of each other node in the same management area, sm ðm – nÞ, to sn . Using the received information, sn calculates the distance dnm between sn and sm as
dnm ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðxm xn Þ2 þ ðym yn Þ2 þ ðzm zn Þ2 :
where t c is the current time, tGPS is the time used to compensate for the error of t nq originating from the location information error of the GPS, and t sync is the time used to compensate for the error of lq ðt c Þ originating from the synchronization error of the internal time between sn and sq . Error of t nq and lq ðtc Þ are systematic errors and can therefore be estimated in advance. The estimation of error of tnq and lq ðt c Þ is described in Section 4.2. If v n ¼ 1, then sn starts to emit sound. If v n ¼ 0, then sn waits for time t pn obtained as follows:
t pn
8 maxðt nm0 þ Dt þ tGPS ðtc lm0 ðt c Þ þ tsync ÞÞ; > > < m0 t c lm0 ðtc Þ þ t sync 6 t nm0 þ t GPS ; ¼ > > : maxðt e 0 þ t nm0 þ Dt þ tGPS Þ; otherwise m0
m
ð25Þ
ð27Þ 0
If dnm < 2R, the service areas of sn and sm overlap. sn registers the coordinate values, priority value, and IP address of sm and flags whether or not the service area is overlapping to the database. The server sends information on sn to sm , which also calculates the distance and registers the information to its database. Through this process, all nodes store information on the other nodes in the same management area in their database. Each node periodically accesses the server to update its database. A node that does not access the server within an arbitrary time limit is deleted from the database of the server and therefore the databases of the other nodes.
where m is the index of the nodes returning 0 in Eq. (26), Dt is a short time margin, and t em0 is the elapsed time until sm0 finishes its emission and sends the end signal to sn . When sn detects the end of the sound event during its emission of sound, it stops its emission and sends an end of emission signal to sm . Then sn increments cn and sets the current time to ln . Through this process, all nodes emit sound in order of priority. cn is set to zero for a suitable interval for any node newly added to the same area. If the waiting time exceeds an arbitrary time limit t lim ; sn starts its emission regardless of the state of other nodes to avoid the excessive delay of its emission.
3.3. Internal time adjustment/synchronization
4. Evaluation of proposed method
The process for controlling the emission timing described below utilizes the internal time of each node, which must be synchronized between nodes having an overlapping service area. Therefore, the internal time of each node has to be absolutely adjusted or relatively synchronized. sn obtains precise time information from its GPS to adjust its internal time. If precise time information cannot be obtained from the GPS owing to bad weather, sn synchronizes its internal clock with other nodes having an overlapping service area using a time synchronization method such as the network time protocol (NTP).
4.1. Delay factors included in public address system employing proposed method
3.4. Transmission of signal and detection of sound event An input signal is fed into one of the nodes in a management area via a microphone embedded in the node or a warning system
In practice, there are delay and error factors with a distribution depending on the configuration of the system. First, the delay factors are investigated in this section. Fig. 7 shows the flows between the input of a signal and the emission of each node. If an input signal is fed into sk ; sk conducts A/D conversion for a duration of sck (s). Then, sk transmits the input signal to node sn for a duration of stkn (s). After receiving the signal, sn detects the start of the sound event for a duration of sdn . Then, sn sends the query to other nodes with an overlapping service area, and on the basis of the responses of these nodes, sn waits a certain time to avoid the overlap of sound. The time from the start of sending the query to the end of waiting is denoted by swn (s). After waiting, sn conducts
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78
75
D/A conversion for a duration of scn (s) to emit the signal. Therefore, the delay between the input of the system and the output from sn , denoted by sn , is expressed as
tlim , then sn terminates the waiting phase and starts D/A conversion to emit sound.
sn ¼ sck þ stkn þ sdn þ swn þ scn :
4.1.3. Delay of sound event detection sdn Delay factor sdn depends on the detection algorithm. As an example, the voice activity detection of ITU-T G.729 Annex B, which utilizes spectral distortion, an energy difference, a low-band energy difference, and a zero-crossing difference, can be completed within 10 ms [13].
4.1.5. Measurements of round-trip time and throughput Delay factors stkn and swn are affected by the throughput and round-trip time, respectively. Therefore, throughput and round-trip-time measurements of the WiFi, LTE, and mesh networks are conducted. Table 1 shows the devices used for the measurements. The software Iperf [14] is used for throughput measurement, and the ping command is used for round-trip-time measurements. In the WiFi measurements, two PCs (Apple Inc. MacBook Pro 2012 Mid) are connected to the WiFi router. In the LTE measurements, two tablets (Google Nexus7 2013 LTE model) are used. One tablet with PCWL-0100 is used for the Iperf server and the other tablet with BM-FRML-1GBM is used for the client. In the mesh network measurements, three access points of the mesh network and two PCs are used. The two PCs and two access points are connected by a 1000BASE-T Ethernet cable. First, a link between the two access points is established. Measurements for this setting are labeled as ‘‘direct’’ in the results. Then, another access point is inserted between the two access points, and links between the three access points are established. Measurements for this setting are labeled as ‘‘1 hop’’ in the results. Table 2 shows the average throughput of each network on the client side over 10 trials. Table 3 shows the average round-trip time of each network over 100 trials. According to the results of the throughput measurements, as an example, if a Pulse Code Modulation (PCM) signal with 16 quantization bits is transmitted via a system employing the proposed method, stkn is below 10 ms when one frame is assumed to consist of 1024 samples.
4.1.4. Delay of waiting time swn Delay factor swn can be expressed as
4.2. Error factors included in public address system employing proposed method
4.1.1. Delay of A/D conversion
ð28Þ
stck and D/A conversion stcn
An input signal is fed into the system via a microphone embedded in each node or a warning system such as J-ALERT. If the input is via an embedded microphone, then delay factor stck depends on the hardware. If the input is via a warning system, it is already digitized and therefore stck ¼ 0. The time required for D/A conversion also depends on the hardware. 4.1.2. Delay of input signal transmission stkn stkn depends on the throughput of the computer network between sn and sk . stkn is expressed as
stkn ¼
s
; n–k b ; 0; otherwise
ð29Þ
where s is the data size per frame and b is the throughput of the computer network.
swn ¼ aðmax ðrtt nm Þ þ bðtpn þ sGPSerr Þ þ cssyncerr Þ; m
ð30Þ
where rtt nm is the round-trip time between sn and sm and a represents how many times sn sends the query to other nodes with an overlapping service area, b represents how many times sn waits a certain time to avoid the overlap of sound on the basis of the responses of other nodes, sGPSerr is an error factor originating from location information obtained from the GPS, c represents how many times the condition tc lm0 ðt c Þ þ t sync 6 t nm0 þ t GPS is satisfied, and ssyncerr is an error factor originating from the synchronization error of the internal time. Even though swn increases with a, if swn exceeds
Not only delay factors but also error factors sGPSerr and ssyncerr appear in the flows shown in Fig. 7. Therefore, the error factors are investigated in this section. 4.2.1. Error of location information The latitude /n , longitude kn , geoid height Ng n , and elevation Hn in Eqs. (22)–(24) obtained from the GPS exhibit variation. These variations result in an error in t nq in Eq. (21), denoted by sGPSerr . Therefore, Eqs. (26) and (27) include t GPS to compensate this error. GPS is normally used with Differential GPS (DGPS) mode utilizes an
Fig. 7. Block diagram of delay and error factors of public address system employing proposed method.
76
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78
Table 1 Devices used for measurements.
Table 4 Standard deviations of location data obtained by GPS.
IEEE 802.11a (WiFi)
Buffalo WAPM-APG300N
LTE
Japan Communications Inc. BM-FRML-1GBM NTT Communication T0003670
IEEE 802.11s (Mesh network)
Sonet International Corp. PCWL-0100
additional standard point to enhance accuracy. There are mainly three type of DGPS: the Japanese Multi-functional Transport Satellite-based Augmentation System (MSAS), the US Wide Area Augmentation System (WAAS) and the European Geostationary Navigation Overlay Service (EGNOS). The designed value of accuracy of these three systems are less than several meters [15–17]. To estimate the order of sGPSerr , the variances of the latitude, longitude, geoid height, and elevation were measured using two GPSs with MSAS: the GPS embedded in Google Nexus7 (2013 LTE model) and Garmin GPS18x LVC. Measurements were conducted on the roof of a building with a clear view of the sky. The measurement period was approximately 3 h, from 3 p.m. to 6 p.m. on June 6, 2014. The sampling rates of these values were 2 Hz for Nexus7 and 1 Hz for GPS18x LVC. The standard deviations of the location data are shown in Table 4. The standard deviations of ECEF coordinate values calculated with the data are shown in Table 5. According to these results and Eqs. (21) and (25), sGPSerr for Nexus7 is of ms order and that for GPS18x LVC is of 10 ms order. 4.2.2. Error of internal time synchronization The most recent finish time of the emissions appearing in Eqs. (26) and (27) has an error originating from the internal time synchronization error between nodes with an overlapping service area, denoted by ssyncerr . Therefore, these equations include the factor t sync to compensate this error. To estimate the order of ssyncerr , measurements of the time synchronization error via each network shown in Table 1 are conducted. In each measurement, ntpd (ver. 4.2.7.p444) is used for time synchronization. The polling interval is set to 8 s. In the LTE measurement, the PC for the ntpd client is connected to the Internet via tethering with Nexus7 and BM-FRML-1GBM. The other PC for the ntpd server is connected to the Internet via a 1000BASE-T Ethernet cable. In the mesh network measurement, the configuration of the connection is the same as that used to measure the throughput and round-trip time. Table 6 shows the offset and jitter of NTP on each network obtained by the ntpq command after synchronization for 30 min. According to the results, ssyncerr for WiFi is of 10 ms order, ssyncerr for LTE is of 100 ms order, and ssyncerr for the mesh network is of ms order. Note that the throughput and round-trip time measurements of LTE are not conducted between PCs and the NTP measurements are conducted under the condition that only the client side PC utilizes LTE. This is because of a limitation of the device that the PC is not allowed to obtain a global IP address using LTE. However, the direct connection between PCs can be achieved by using a particular plan provided by a carrier or VPN even for LTE. Therefore, LTE can be used in the proposed method. Table 2 Throughput of each network (Mbps). Average WiFi LTE Mesh network(direct) Mesh network(1 hop)
27.11 1.962 13.50 5.070
/ (deg.)
k (deg.)
Nexus7
1:0289 10
GPS18x LVC
8:2375 106
6
3:3536 10
Ng (m) 7
8:8742 106
H (m)
0
0.60287
5:5390 1012
7.7818
Table 5 Standard deviations of coordinate values calculated with data obtained from GPS (m).
Nexus7 GPS18x LVC
x
y
z
0.39127 4.0248
0.41518 5.5292
0.22824 3.9141
4.3. Signal to interference ratio with/without proposed method In this section, the performance of the proposed method is evaluated in terms of the signal to interference ratio. The signal to interference ratio of a public address system with and without the proposed method is estimated as follows. In this estimation, N nodes are assumed to form a regular polygon with overlapping service areas. The example of N ¼ 3 is shown in Fig. 8. Listening points are located in the service areas of n ðn 6 NÞ nodes, and listening points in overlapping service areas are equidistant from the neighboring nodes. This arrangement of nodes is a rare case, however, it is employed to simulate worse-case scenario. The signal to interference ratio is estimated using Eq. (7) by considering the sound from one node as the signal and the sounds from other nodes as interference. In Fig. 8, for example, the numbers of sounds acting as interference are 0 at listening point A, 1 at B, and 2 at C. The gain factor of each node is assumed to be 1. The impulse responses between nodes and listening points are assumed to have attenuation and delay factors that depend only on the distance. Delay factors mentioned in Section 4.1 other than the delays of D/A conversion do not produce sound overlap in the proposed method. The amount of sound overlap resulted from the delay of D/A conversion is negligible small. On the other hand, error factors measurements in Section 4.2 have the possibility to produce sound overlap as shown in Fig. 9. According to the results of error factors measurements in Section 4.2, the maximum amount of sound overlap resulted from error factors can be considered to be below 200 ms in the proposed method. The duration of the input signal is set to 3 s in this estimation. The absolute value of the amplitude of the input signal is consistently set to 1 over its duration to simulate severe condition. The ratio of the period with SIR = 0 dB to the emission period is calculated with 200 ms of the amount of sound overlap resulted from error factors. Fig. 10 shows the average of this ratio over each node. Without the proposed method, each node emits sound simultaneously, therefore the ratio of the period with SIR = 0 dB to the emission period is consistently 100%. With the proposed method, the amount of sound overlap resulted from error factors increases with increasing the number of interference nodes. However, the ratio with the proposed method is consistently less than the ratio without proposed method. Moreover, because the setting of this estimation is worse-case scenario, the actual impact of error factors to SIR can be considered smaller.
Table 3 Round-trip time of each network (ms). Standard deviation 0.8439 0.5010 0.9920 0.4770
WiFi LTE Mesh network(direct) Mesh network(1 hop)
Average
Standard deviation
58.71 237.4 2.170 3.684
30.93 162.4 0.130 0.409
77
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78 Table 6 Offset and jitter of NTP on each network (ms).
WiFi LTE Mesh network (direct) Mesh network (1 hop)
S3
Offset
Jitter
1.573 220.19 0.558 0.951
28.846 256.934 0.063 0.122
Sound overlap resulted from error factors
S2 SIR2 = +∞
SIR2 = 0
SIR2 = 0
S1
t
5. Demonstration on laptop PC
Emission timing difference resulted from error factors Fig. 9. Example of sound overlap due to error factors.
Raito of period when SIR = 0 dB
100
80
60
40
20
0
0
1
2
3
4
5
Number of interference nodes
Fig. 10. The average ratio of period when SIR = 0 dB of a public address system with/without proposed method.
s3
Index of node
The proposed method is implemented on a laptop PC (Apple Inc. MacBook Pro 2012 Mid). Three nodes, s1 ; s2 , and s3 , are separately launched on three PCs connected to each other via WiFi using the devices listed in Table 1. A server to store the information on the nodes in one area is also launched on another PC. The internal times of the three PCs are preliminarily synchronized using ntpd (ver. 4.2.7p444) instead of a GPS to simulate the worse-case scenario. s2 is set to be the NTP server and s1 and s3 are set to be the NTP client to obtain time information on s2 . The offset and jitter of s1 and s3 are 11.893 ms and 23.816 ms, and 2.859 ms and 33.509 ms, respectively. The coordinate values of the three nodes, (x1 ; y1 ; z1 ), (x2 ; y2 ; z2 ), and (x3 ; y3 ; z3 ), are determined so that the three nodes form a regular triangle such as the one shown in Fig. 8, and the distance between each node is 400 m. Then, the standard deviations of the coordinate values, calculated using the location information obtained by Nexus7 shown in Table 5, are added to the coordinate values of each node. t GPS is set to 10 ms in accordance with the estimation in Section 4.2.1 and tsync is set to 50 ms in accordance with the estimation in Section 4.2.2. The diameter of the service area of each node R is set to 300 m, the height of the loudspeakers H is set to 15 m, the height of the listening points h is set to 1.5 m, and the speed of sound c is set to 340 m/s. The priority value of each node is set so that p1 < p2 < p3 ; therefore, the nodes are expected to emit sound in the order s1 ; s2 ; s3 . After each node finishes the construction of the database containing information on the other nodes, the input signal is fed into s1 . The input signal is a pink noise having 3 s duration with 1 s silence added before and after its duration. The sampling frequency of the input signal is 44100 Hz and the number of quantization bits is 16. In this implementation, 10 frames are used to detect the start of the sound event and three frames are used to detect the end of the sound event. Fig. 11 shows the emission timing of each node. According to Eq. (21), t12 ¼ t23 ¼ 0:586 s. In this trial, the time between the end of emission of s1 and the start of emission of s2 is 0.789 s and the time between the end of emission
s2
s1
0
2
4
6
8
10
12
Time (s) Fig. 11. Emission timing of demonstration system.
of s2 and the start of emission of s3 is 0.859 s. Both these elapsed times exceed t 12 and t 23 . Therefore, the sounds emitted from each node do not overlap at listening points within the service areas of these nodes under the assumptions that the attenuation and delay factor between a loudspeaker and a listening point depends only on the distance and the directivity of each node is omnidirection. 6. Conclusions
Fig. 8. Nodes with overlapping service areas and listening points.
In this paper, to minimize the sound overlap of a public address system for emergency announcements, a method for controlling
78
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78
the timing of emission was proposed. The proposed method focuses on the input and output of a public address system. Under the assumptions that the attenuation and delay factor between a loudspeaker and a listening point depends only on the distance and the directivity of each node is assumed to be omnidirection, the optimal emission timing for a pair of nodes is expressed in terms of the signal to interference ratio. Then, on the basis of the above optimal emission timing, the emission timing for a set of nodes in a management area is determined. In the proposed method, each node collects and shares the positions of each node obtained by GPS via a computer network to determine the emission timing, and then taking into account whether other nodes emit sound or not each node autonomously determines its emission timing and the delay time for other nodes. The delay and error factors included in a public address system employing the proposed method are also modeled and estimated. The proposed method can be implemented without any dedicated hardware, as shown in Section 5, and, therefore, can be applied to existing loudspeakers. In future work, more detailed consideration of the signal transmission between a loudspeaker and a listening point and the directivity of node is needed. In particular, the service area of each node is assumed to be a perfect circle in the proposed method. Therefore, a method of controlling the emission level to compensate for the distortion of the service area caused by wind, rainfall, and reflection from buildings should be developed. Acknowledgments Part of this work was carried out with the support of a Grant-in-Aid for the Strategic Information and Communications R&D Promotion Programme (SCOPE) of the Ministry of Internal Affairs and Communications of Japan No. 132310012 ‘‘Intelligent system for public address in the open air with mesh network for transmitting disaster mitigation information’’. References [1] NTT DOCOMO, INC. Area mail disaster information service,; 2010 [accessed 6.03.15].
[3] Material of the Committee for Technical Investigation on Countermeasures for Earthquakes and Tsunamis Based on the Lessons Learned from the ‘‘2011 off the Pacific coast of Tohoku Earthquake,’’, 2011 [in Japanese, accessed 6.03.15]. [4] Report of the Committee for Technical Investigation on Countermeasures for Earthquakes and Tsunamis Based on the Lessons Learned from the ‘‘2011 off the Pacific coast of Tohoku Earthquake,’’ ; 2011 [accessed 6.03.15]. [5] Saito F, Cui Z, Sato H, Morimoto M, Chisaki Y, Usagawa T, Iwaya Y, Sakamoto S, Suzuki Y, Aoki M, Takashima K. Toward advanced open-air PA systems to convey emergency disaster information. In: Proc. autumn meeting of the Acoustical Society of Japan; 2012. p. 725–6 [in Japanese]. [6] Mori J, Satoh F, Yokoyama S, Tachibana H. Prediction of outdoor sound propagation by geometrical computer modeling. Acoust Sci Technol 2014;35(1):50–4. [7] Sato H, Cui Z, Sakamoto S, Suzuki Y, Morimoto M, Aoki M, Koike H, Takashima K, Tsuru H, Mitsueda T. Evaluation of outdoor public address system based on speech intelligibility. In: Proc. autumn meeting of the Acoustical Society of Japan; 2013. p. 1533–6 [in Japanese]. [8] Miyagawa Y, Sato H, Morimoto M, Suzuki Y, Sakamoto S. Relationship between sentence intelligibility and acoustic objective measures in sound fields with a long-path echo. In: Proc. spring meeting of the Acoustical Society of Japan; 2014. p. 1547–50 [in Japanese]. [9] Ministry of International Affairs and Communications. The radio use website — administrative radio system for disaster use, [accessed 6.03.15]. [10] Kawase Y, Takayama T, Shinohara N. Survey of the relationship of level fluctuations of outdoor public emergency broadcasting sound with meteorological conditions. In: Proc. autumn meeting of the Institute of Noise Control Engineering of Japan; 2012. p. 213–6 [in Japanese]. [11] Okubo T, Yokota T, Makino K, Ohshima T, Okada Y, Hiraguri Y, Kawase Y, Imaizumi H. Meteorological effect on outdoor propagation of emergency public addressing. In: Proc. autumn meeting of the Acoustical Society of Japan; 2013. p. 1541–4 [in Japanese]. [12] Ministry of Internal Affairs and Communications. 2012 white paper on information and communication in Japan, promotion of computerization in the fire safety and disaster preparedness field, , [in Japanese, with a schematic diagram], ; 2012 [accessed 6.03.15]. [13] ITU-T. A silence compression scheme for G.729 optimized for terminals conforming to recommendation v.70, Recommendation G.729 Annex B, International Telecommunication Union; 1996. [14] Iperf. ; 2012 [in Japanese, accessed 6.03.15]. [16] Satellite Navigation – WAAS – Benefits, [accessed 6.03.15].
Contents lists available at ScienceDirect
Applied Acoustics journal homepage: www.elsevier.com/locate/apacoust
Emission timing control method for improving signal to interference ratio on public address system Taira Onoguchi ⇑, Dan Murakami, Yoshifumi Chisaki Graduate School of Science and Technology, Kumamoto University, 2-39-1 Kurokami, Chuo-ku, Kumamoto 860-8555, Japan
a r t i c l e
i n f o
Article history: Received 11 September 2014 Received in revised form 20 March 2015 Accepted 30 April 2015 Available online 18 May 2015 Keywords: Public address system Sound overlap Emission timing control Computer network Disaster mitigation
a b s t r a c t In order to reduce disaster damage, there are several media such as email, TV, radio and a public address system with outdoor loudspeakers to deliver emergency announcements to the people at risk. The Great East Japan Earthquake revealed the importance of a public address system with outdoor loudspeakers for emergency announcements and the difficulty of its intelligibility. To achieve a public address system with sufficient intelligibility for residents, one important factor is to suppress sound overlap caused by reflection from terrain or buildings, or caused by the simultaneous emission of other loudspeakers. Recent Internet technologies and devices will contribute to avoid sound overlap caused by the simultaneous emission of other loudspeakers. In this paper, a method of controlling emission timing to minimize the sound overlap caused by the simultaneous emission of other loudspeakers of a public address system is proposed. This method focused on the transmission of a signal between the input and output of a public address system. Under the assumptions that the attenuation and delay factor between a loudspeaker and a listening point depend only on the distance and the directivity of each node is omnidirection, the optimal emission timing for a pair of nodes is expressed in terms of the signal to interference ratio. Then, on the basis of the above optimal emission timing, the emission timing for a set of nodes in a management area is determined. In the proposed method, each node collects and shares the positions of each node obtained by the Global Positioning System via a computer network to determine the emission timing, and then taking into account whether other nodes emit sound or not each node autonomously determines its emission timing and the delay time for other nodes. The proposed method can be implemented without any dedicated hardware and can, therefore, be applied to existing loudspeakers. The delay and error factors in a public address system employing the proposed method are also modeled and estimated. Ó 2015 Elsevier Ltd. All rights reserved.
1. Introduction In order to reduce disaster damage, there are several media such as email, TV, radio [1,2] and a public address system with outdoor loudspeakers shown in Fig. 1 to deliver emergency announcements to the people at risk. In the Great East Japan Earthquake, among residents who could obtain evacuation announcements in the affected prefectures, 45% of them obtained it from outdoor loudspeakers. However, among those who obtained evacuation announcements from outdoor loudspeakers, 56% of them answered they could clearly hear the announcements [3]. Therefore, the Committee for Technical Investigation on Countermeasures for Earthquakes and Tsunamis Based on the Lessons Learned from the ‘‘2011 off the Pacific coast of Tohoku Earthquake’’ suggests improvement of ⇑ Corresponding author. Tel.: +81 96 3423621; fax: +81 96 3423630. E-mail addresses: [email protected] (T. Onoguchi), chisaki@cs. kumamoto-u.ac.jp (Y. Chisaki). http://dx.doi.org/10.1016/j.apacoust.2015.04.019 0003-682X/Ó 2015 Elsevier Ltd. All rights reserved.
intelligibility of a public address system as one of the countermeasures to mitigate tsunami damage [4]. To achieve addressing with sufficient intelligibility for residents, factors that degrade the intelligibility of radiated sound must be minimized. One important factor is sound overlap at observation points. In particular, sound overlap caused by reflection from terrain or buildings, or caused by the simultaneous emission of other loudspeakers, is known to degrade intelligibility [5–7]. As demonstrated by the hearing ability of humans, and illustrated by the cocktail party effect, intelligibility depends not only on the power ratio between the target signal and interference but also on the context of the speech, the characteristics of the signal, the direction of arrival, and word familiarity. However, in the previous work it was shown to be possible for intelligibility to be estimated from the power ratio of overlapping sounds [8]. In accordance with the result of this previous work, in this paper, sound overlap is defined by the power ratio of the signal arrived from one node to the signals arrived from other nodes. Therefore, in this paper, the impact of the context of the speech,
71
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78
This paper is organized as follows. In Section 2, the optimal emission timing for a pair of nodes in terms of the signal to interference ratio is derived. Then a model for determining the emission timing for a set of nodes in a management area is shown. In Section 3, to accomplish the optimal emission timing derived in Section 2, a method of controlling emission timing for a public address system is proposed. In Section 4, the delay and error factors in a public address system employing the proposed method are investigated. In Section 5, the proposed method is implemented on a laptop PC without any dedicated hardware and a demonstration is conducted. Conclusions are given in Section 6. 2. Model for determining emission timing for public address system 2.1. Signal to interference ratio in public address system
Fig. 1. Outdoor loudspeaker of a public address system for emergency use.
the characteristics of the signal, the direction of arrival, and word familiarity are not considered. Because the signals emitted by the nodes of a public address system in one area are the same, when the phase of overlapped sounds is the same, the sound is emphasized rather than masked. However, in the actual situation, this kind of phase synchronization can be considered as a rare case. Therefore, in this paper, the impact of phase is not considered. Based on this definition of sound overlap, sound arriving from adjacent loudspeakers can be considered to degrade intelligibility. Therefore, a method of suppressing sound overlap between adjacent loudspeakers is required. One simple way is to set loudspeakers so that each service area of them does not have overlap. However, since there are loudspeakers already installed, it costs so much to move them. For example, at the end of 2001, 65% of all the local governments of cities, towns, and villages in Japan was equipped with the broadcast system for disaster [9]. In this paper, a method for controlling the timing of emission to reduce sound overlap in a public address system is proposed. In a public address system with loudspeakers, an input signal is transmitted via a path consisting of two portions. The first portion is the path between the input and output of the public address system. The second portion is the path between a loudspeaker and a listening point. The latter path is affected by many factors such as rainfall, absorption by air and terrain, the distribution of temperature, humidity, reflection by terrain and buildings, and wind. In particular, previous works showed that open-air sound transmission between a loudspeaker in a public address system and a listening point are affected by reflection [6] and wind [10,11]. In this paper, the latter path is simplified such that attenuation and delay factors of the impulse response between a loudspeaker and a listening point only depend on the distance. The proposed method focuses on controlling the emission timing for the former path, the path between the input and output of the public address system. The proposed method delays start time of emission of each node to avoid sound overlap by following processes. Each node collects and shares the positions of each node obtained by the Global Positioning System (GPS) via a computer network to determine the emission timing. An input signal can be fed into one of each node in the same management area. The input signal is transmitted to other nodes, then each node detects the start time of a sound event from the transmitted signal. After detecting the start of a sound event, a node sends a query to other nodes to obtain their status of emission. On the basis of response from other nodes, each node autonomously determines its emission timing.
The amplitude of an input signal fed into a public address system at time t is denoted by
aðtÞ ¼
A; 0 6 t 6 N 0;
otherwise
;
ð1Þ
where A 2 R and N is the duration of the input signal. The amplitude of the output signal of node sn at t, denoted by bn ðtÞ, is obtained as
bn ðtÞ ¼ g n aðt sn Þ;
ð2Þ
where g n is the gain factor and sn is the delay factor of sn . Then the amplitude of an observed signal at a listening point in the service area of sn at time t, denoted by yn ðtÞ, is calculated as
yn ðtÞ ¼ ¼
1 X
bn ðkÞhn ðt kÞ;
ð3Þ
g n aðk sn Þhn ðt kÞ;
ð4Þ
k¼1 1 X k¼1
where hn ðtÞ is the impulse response between sn and the listening point. yn ðtÞ can be rewritten as
yn ðtÞ ¼ g n an aðt sn bn Þ;
ð5Þ
where an is the attenuation factor and bn is the delay factor of hn ðtÞ. an and bn depend on many factors such as distance attenuation, rainfall, wind, reflection by terrain and buildings, absorption by air and terrain, the distribution of temperature, and humidity. In this paper, an and bn are assumed to depend only on the distance as follows:
rn yn ðtÞ ¼ g n aðr n Þa t sn ; c
ð6Þ
where r n is the distance between sn and a listening point and c is the speed of sound. Therefore, the signal to interference ratio at time t regarding the sound from sn as the signal and the sound from other nodes sm ðm – nÞ as interference, denoted by SIRn ðtÞ, can be obtained as follows:
jy ðtÞj SIRn ðtÞ ¼ 20log10 P n ; m jym ðtÞj g aðr n Þjaðt sn rcn Þj ¼ 20log10 P n rm : m g m aðr m Þjaðt sm c Þj
ð7Þ ð8Þ
The vector of the optimal delay factor for each node in terms of the signal to interference ratio, sopt , is then represented as
XXX SIRn ðtÞ;
sopt ¼ argmax sn ;sm
n
ð9Þ
S t2tn
P where tn is the vector of time when yn ðtÞ – 0 and S is the summation over listening points in the service area of sn .
72
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78
2.2. Optimal emission timing for pair of nodes in terms of signal to interference ratio For a pair of nodes sa and sb with the delay factor satisfying sopt , the following condition is required:
rb a t sb ¼ 0; c
ra if a t sa – 0: c
ð10Þ
Because aðtÞ ¼ 0 for t > N according to Eq. (1), this requirement can be converted to the following equation:
sa þ
ra rb > sb þ þ N: c c
ð11Þ
Therefore,
sa > sb þ
rb ra þ N: c
ð12Þ
It is sufficient to satisfy Eq. (12) only at the listening point on the near-side edge of sb for sa ; Ladj . The position of Ladj depends on the pattern of overlapping service areas. Fig. 2 shows patterns of overlapping service areas. In this paper, the directivity of each node is assumed to be omnidirection for simplifying. Therefore, with the assumption that an depends only on the distance, the service area of each node is assumed to be a perfect circle. Hereinafter, the diameter of the service area of each node is denoted by R and the distance between sa and sb is denoted by dab .
2.3. Determination of emission timing for set of nodes in management area Node sn is represented by
sn ¼ ½ pn
cn ðtÞ en ðtÞ ln ðtÞ ;
ð14Þ
where pn is the value for the priority of sn . Priority is preliminarily given to each node taking factors such as population into account. Duplicate priority values are not allowed in the same management area. Also, a node located near the border of a management area must not have a priority value that is duplicated in each adjacent area. cn ðtÞ is the emission count of sn up to time t. en ðtÞ is a flag that takes a value of 1 when sn emits a sound at time t and 0 when sn does not emit a sound at t. ln ðtÞ is the most recent finish time of the emissions of sn . Hereinafter, Sn represents the service area of sn and Om represents an overlapping service area. If there are N nodes and M service areas overlapping in a management area, matrix B is defined by
2
b11 6 . 6 . 6 . 6 B¼6 6 bn1 6 . 6 . 4 . bN1
b1m .. .
bnm .. .
bNm
3 b1M .. 7 7 . 7 7 bnM 7 7; .. 7 7 . 5
bnm ¼
1; Sn Om 0;
otherwise
:
ð15Þ
bNM
The situation in Fig. 3, for example, can be represented as
(1) Pattern A (dab < R). qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 2 At Ladj , r a ¼ ðR dab Þ þ ðH hÞ and r b ¼ R2 þ ðH hÞ ,
ð16Þ
where H is the height of the loudspeaker and h is the height of the listening point. Therefore, Eq. (12) is converted to
sa > sb þ N
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 2 R2 þ ðH hÞ ðR dab Þ þ ðH hÞ
Matrix O is defined by
ð13Þ
2
(2) Pattern B (R 6 dab 6 2R). qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 2 At Ladj , r a ¼ ðdab RÞ þ ðH hÞ and rb ¼ R2 þ ðH hÞ .
6 6 6 O¼6 6 4
þ
c
:
Therefore, Eq. (12) is again converted to Eq. (13). (3) Pattern C (dab > 2R). There are no service areas overlapping. Therefore, no requirement is derived from Eq. (12).
Pattern A
Pattern B
Pattern C
sα
sβ
O1
0
0 .. . 0
O2
3 0 .. 7 . 7 7 7: 7 0 5
..
. 0
ð17Þ
OM
Then matrix U is defined as
2u 3 1
6 .. 7 6 . 7 6 7 6 7 U ¼ 6 un 7 ¼ BO: 6 7 6 .. 7 4 . 5
ð18Þ
uN
Ladj
: Service area Ladj
2
Ladj rα
4
1
rβ
H 2
h dαβ
3
1
R
Fig. 2. Patterns of overlapping service areas.
Fig. 3. Example of overlapping service areas.
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78
The flag indicating whether sn can emit sound at t; v n ðtÞ, can be obtained from matrix U and operator VðtÞ as
v n ðtÞ ¼ VðtÞun ¼
1; 8k 2 k; Gðt; Ok Þ ¼ 1; or k ¼ ; ; 0; otherwise
ð19Þ
where k is a vector consisting of the indexes of overlapping service areas included in un . If Ok ¼ Sn \ Sq , the operator G returns a value such that
8 1; eq ðtÞ – 1; and t lq ðtÞ > tnq ; and > < p pn þ cn ðtÞ < maxðpqn ;pq Þ þ cq ðtÞ ; Gðt; Ok Þ ¼ maxðpn ;pq Þ > : 0; otherwise
ð20Þ
where tnq is the required difference in emission timing between sn and sq to avoid the overlap of emitted sounds. The above conditional equation allows a node with a smaller priority value to emit sound earlier. According to Eq. (13), t nq is obtained as
t nq ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 2 R2 þ ðH hÞ ðR dnq Þ þ ðH hÞ c
ð21Þ
:
3. Method of controlling emission timing for public address system To accomplish emission timing control based on the emission timing determined in Section 2.3 in accordance with Eqs. (20) and (21), each node must obtain the following information. Duration of input signal. Distance between nodes. Priority, emission count, the flag indicating whether or not emission is being conducted now, and the most recent finish time of the emission of other nodes with which the service area overlaps. Precise time information. Because a signal transmitted over a public address system is initially unknown, each node must utilize some method to detect the start and end of the sound event to determine the duration of the input signal. To obtain the distance between nodes, each node must obtain information on its location. Also, the information listed above must be shared among the nodes in a management area. A method of controlling the emission timing for a public address system that meets these requirements is proposed in this section. A schematic diagram of the proposed method is shown in Fig. 4. A block diagram and a flowchart of the proposed method are shown in Figs. 5 and 6, respectively. The proposed method utilizes one server and N nodes, s1 ; s2 ; . . . ; sn ; . . . ; sN , for a management area. A server consists of a database containing information on the nodes in the same management area and network devices used to exchange information with nodes. Each node consists of loudspeakers, a microphone to input the signal, a receiver for warning Server of each management area
systems such as the national early warning system (J-ALERT) [12], a database containing information on other nodes, a buffer to store input signals, a GPS to obtain its location and precise time information, and network devices to exchange information with the server and other nodes. The priority value pn is preliminarily given to sn taking factors such as the local population into account. Duplicate priority values are not allowed in the same management area. A node located near the border of a management area must not have a priority value that is duplicated in each adjacent area. Under the assumptions that the attenuation and delay factor between a loudspeaker and a listening point depend only on the distance and the directivity of each node is omnidirection, the service area of each node is assumed to be a perfect circle with diameter R. 3.1. Selection of active network for information exchange Each node can select one active network from several types of available network, such as WiFi, 3rd generation (3G), Long Term Evolution (LTE), or a mesh network, for backup in the case of a network being down. The selected network is used for information exchange between the server and nodes and between nodes. 3.2. Collection of own information by node and construction of database of other nodes Node sn obtains information on its own location from a GPS or preliminary measurements. This information consists of the latitude /n , longitude kn , geoid height Hgn , and elevation En . These values are converted to earth-centered earth-fixed (ECEF) coordinate values xn ; yn , and zn as follows:
xn ¼ ðN þ En þ Hgn Þ cos /n cos kn ;
Network devices GPS
ð23Þ
zn ¼ ðNð1 e2 Þ þ En þ Hgn Þ sin /n ;
ð24Þ
where N is the radius of the prime vertical and e is the first eccentricity. After calculating the coordinate values, sn sends them and its priority value pn to the server in the same management area. If node sn is located near the border of another management area, it sends
Network devices
Receiver for warning systems Node
Network devices GPS
Mic.
Receiver for warning systems Node
Fig. 4. Schematic diagram of proposed method.
ð22Þ
yn ¼ ðN þ En þ Hgn Þ cos /n sin kn ;
WiFi, LTE, 3G, mesh network
Mic.
73
Fig. 5. Block diagram of proposed method.
74
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78
such as J-ALERT. Then the input signal is transmitted to other nodes and stored in the buffer of each node. Each node reads the signal from its buffer. If it detects the start of a sound event, it starts the following process to determine the emission timing.
Read input signals from buffer Start of sound event is detected ?
No
3.5. Determination and control of emission timing
Yes
Send query to other nodes which overlap service area
vn = 1 ?
No
Wait tp (s)
After detecting the start of a sound event, node sn sends a query to other nodes with an overlapping service area, sm ðm – nÞ, to obtain the current values of pm ; cm ; em , and lm described in Section 2.3. If sn does not obtain a response from a node within an arbitrary time limit, this node is considered as being down and its state is ignored in the determination of the emission timing. Depending on the response from sm , Eq. (19) is evaluated. In this process, owing to error factors, the following equation is used instead of Eq. (20):
n
Exceeds No
tlim ?
Yes
Yes
Emit signals Yes
End of sound event is detected ?
No
8 1; eq ðt c Þ – 1; and t c lq ðtc Þ þ t sync > tnq þ t GPS ; and > < p pn þ cn ðtc Þ < maxðpq ;p Þ þ cq ðt c Þ Gðtc ;Ok Þ ¼ ; maxðpn ;pq Þ n q > : 0; otherwise ð26Þ
Yes
Send end signal to other nodes Fig. 6. Flow chart of proposed method. v n is the flag of emission timing determination, tpn is the waiting time until resending of a query, and tlim is an arbitrary time limit for waiting.
the coordinate values and priority value to each server in the adjacent area. After receiving this information, the server registers the coordinate values, priority value, and IP address of sn with a time stamp to the database. Then, the server sends the coordinate values xm ; ym , and zm , the priority value pm , and the IP address of each other node in the same management area, sm ðm – nÞ, to sn . Using the received information, sn calculates the distance dnm between sn and sm as
dnm ¼
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðxm xn Þ2 þ ðym yn Þ2 þ ðzm zn Þ2 :
where t c is the current time, tGPS is the time used to compensate for the error of t nq originating from the location information error of the GPS, and t sync is the time used to compensate for the error of lq ðt c Þ originating from the synchronization error of the internal time between sn and sq . Error of t nq and lq ðtc Þ are systematic errors and can therefore be estimated in advance. The estimation of error of tnq and lq ðt c Þ is described in Section 4.2. If v n ¼ 1, then sn starts to emit sound. If v n ¼ 0, then sn waits for time t pn obtained as follows:
t pn
8 maxðt nm0 þ Dt þ tGPS ðtc lm0 ðt c Þ þ tsync ÞÞ; > > < m0 t c lm0 ðtc Þ þ t sync 6 t nm0 þ t GPS ; ¼ > > : maxðt e 0 þ t nm0 þ Dt þ tGPS Þ; otherwise m0
m
ð25Þ
ð27Þ 0
If dnm < 2R, the service areas of sn and sm overlap. sn registers the coordinate values, priority value, and IP address of sm and flags whether or not the service area is overlapping to the database. The server sends information on sn to sm , which also calculates the distance and registers the information to its database. Through this process, all nodes store information on the other nodes in the same management area in their database. Each node periodically accesses the server to update its database. A node that does not access the server within an arbitrary time limit is deleted from the database of the server and therefore the databases of the other nodes.
where m is the index of the nodes returning 0 in Eq. (26), Dt is a short time margin, and t em0 is the elapsed time until sm0 finishes its emission and sends the end signal to sn . When sn detects the end of the sound event during its emission of sound, it stops its emission and sends an end of emission signal to sm . Then sn increments cn and sets the current time to ln . Through this process, all nodes emit sound in order of priority. cn is set to zero for a suitable interval for any node newly added to the same area. If the waiting time exceeds an arbitrary time limit t lim ; sn starts its emission regardless of the state of other nodes to avoid the excessive delay of its emission.
3.3. Internal time adjustment/synchronization
4. Evaluation of proposed method
The process for controlling the emission timing described below utilizes the internal time of each node, which must be synchronized between nodes having an overlapping service area. Therefore, the internal time of each node has to be absolutely adjusted or relatively synchronized. sn obtains precise time information from its GPS to adjust its internal time. If precise time information cannot be obtained from the GPS owing to bad weather, sn synchronizes its internal clock with other nodes having an overlapping service area using a time synchronization method such as the network time protocol (NTP).
4.1. Delay factors included in public address system employing proposed method
3.4. Transmission of signal and detection of sound event An input signal is fed into one of the nodes in a management area via a microphone embedded in the node or a warning system
In practice, there are delay and error factors with a distribution depending on the configuration of the system. First, the delay factors are investigated in this section. Fig. 7 shows the flows between the input of a signal and the emission of each node. If an input signal is fed into sk ; sk conducts A/D conversion for a duration of sck (s). Then, sk transmits the input signal to node sn for a duration of stkn (s). After receiving the signal, sn detects the start of the sound event for a duration of sdn . Then, sn sends the query to other nodes with an overlapping service area, and on the basis of the responses of these nodes, sn waits a certain time to avoid the overlap of sound. The time from the start of sending the query to the end of waiting is denoted by swn (s). After waiting, sn conducts
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78
75
D/A conversion for a duration of scn (s) to emit the signal. Therefore, the delay between the input of the system and the output from sn , denoted by sn , is expressed as
tlim , then sn terminates the waiting phase and starts D/A conversion to emit sound.
sn ¼ sck þ stkn þ sdn þ swn þ scn :
4.1.3. Delay of sound event detection sdn Delay factor sdn depends on the detection algorithm. As an example, the voice activity detection of ITU-T G.729 Annex B, which utilizes spectral distortion, an energy difference, a low-band energy difference, and a zero-crossing difference, can be completed within 10 ms [13].
4.1.5. Measurements of round-trip time and throughput Delay factors stkn and swn are affected by the throughput and round-trip time, respectively. Therefore, throughput and round-trip-time measurements of the WiFi, LTE, and mesh networks are conducted. Table 1 shows the devices used for the measurements. The software Iperf [14] is used for throughput measurement, and the ping command is used for round-trip-time measurements. In the WiFi measurements, two PCs (Apple Inc. MacBook Pro 2012 Mid) are connected to the WiFi router. In the LTE measurements, two tablets (Google Nexus7 2013 LTE model) are used. One tablet with PCWL-0100 is used for the Iperf server and the other tablet with BM-FRML-1GBM is used for the client. In the mesh network measurements, three access points of the mesh network and two PCs are used. The two PCs and two access points are connected by a 1000BASE-T Ethernet cable. First, a link between the two access points is established. Measurements for this setting are labeled as ‘‘direct’’ in the results. Then, another access point is inserted between the two access points, and links between the three access points are established. Measurements for this setting are labeled as ‘‘1 hop’’ in the results. Table 2 shows the average throughput of each network on the client side over 10 trials. Table 3 shows the average round-trip time of each network over 100 trials. According to the results of the throughput measurements, as an example, if a Pulse Code Modulation (PCM) signal with 16 quantization bits is transmitted via a system employing the proposed method, stkn is below 10 ms when one frame is assumed to consist of 1024 samples.
4.1.4. Delay of waiting time swn Delay factor swn can be expressed as
4.2. Error factors included in public address system employing proposed method
4.1.1. Delay of A/D conversion
ð28Þ
stck and D/A conversion stcn
An input signal is fed into the system via a microphone embedded in each node or a warning system such as J-ALERT. If the input is via an embedded microphone, then delay factor stck depends on the hardware. If the input is via a warning system, it is already digitized and therefore stck ¼ 0. The time required for D/A conversion also depends on the hardware. 4.1.2. Delay of input signal transmission stkn stkn depends on the throughput of the computer network between sn and sk . stkn is expressed as
stkn ¼
s
; n–k b ; 0; otherwise
ð29Þ
where s is the data size per frame and b is the throughput of the computer network.
swn ¼ aðmax ðrtt nm Þ þ bðtpn þ sGPSerr Þ þ cssyncerr Þ; m
ð30Þ
where rtt nm is the round-trip time between sn and sm and a represents how many times sn sends the query to other nodes with an overlapping service area, b represents how many times sn waits a certain time to avoid the overlap of sound on the basis of the responses of other nodes, sGPSerr is an error factor originating from location information obtained from the GPS, c represents how many times the condition tc lm0 ðt c Þ þ t sync 6 t nm0 þ t GPS is satisfied, and ssyncerr is an error factor originating from the synchronization error of the internal time. Even though swn increases with a, if swn exceeds
Not only delay factors but also error factors sGPSerr and ssyncerr appear in the flows shown in Fig. 7. Therefore, the error factors are investigated in this section. 4.2.1. Error of location information The latitude /n , longitude kn , geoid height Ng n , and elevation Hn in Eqs. (22)–(24) obtained from the GPS exhibit variation. These variations result in an error in t nq in Eq. (21), denoted by sGPSerr . Therefore, Eqs. (26) and (27) include t GPS to compensate this error. GPS is normally used with Differential GPS (DGPS) mode utilizes an
Fig. 7. Block diagram of delay and error factors of public address system employing proposed method.
76
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78
Table 1 Devices used for measurements.
Table 4 Standard deviations of location data obtained by GPS.
IEEE 802.11a (WiFi)
Buffalo WAPM-APG300N
LTE
Japan Communications Inc. BM-FRML-1GBM NTT Communication T0003670
IEEE 802.11s (Mesh network)
Sonet International Corp. PCWL-0100
additional standard point to enhance accuracy. There are mainly three type of DGPS: the Japanese Multi-functional Transport Satellite-based Augmentation System (MSAS), the US Wide Area Augmentation System (WAAS) and the European Geostationary Navigation Overlay Service (EGNOS). The designed value of accuracy of these three systems are less than several meters [15–17]. To estimate the order of sGPSerr , the variances of the latitude, longitude, geoid height, and elevation were measured using two GPSs with MSAS: the GPS embedded in Google Nexus7 (2013 LTE model) and Garmin GPS18x LVC. Measurements were conducted on the roof of a building with a clear view of the sky. The measurement period was approximately 3 h, from 3 p.m. to 6 p.m. on June 6, 2014. The sampling rates of these values were 2 Hz for Nexus7 and 1 Hz for GPS18x LVC. The standard deviations of the location data are shown in Table 4. The standard deviations of ECEF coordinate values calculated with the data are shown in Table 5. According to these results and Eqs. (21) and (25), sGPSerr for Nexus7 is of ms order and that for GPS18x LVC is of 10 ms order. 4.2.2. Error of internal time synchronization The most recent finish time of the emissions appearing in Eqs. (26) and (27) has an error originating from the internal time synchronization error between nodes with an overlapping service area, denoted by ssyncerr . Therefore, these equations include the factor t sync to compensate this error. To estimate the order of ssyncerr , measurements of the time synchronization error via each network shown in Table 1 are conducted. In each measurement, ntpd (ver. 4.2.7.p444) is used for time synchronization. The polling interval is set to 8 s. In the LTE measurement, the PC for the ntpd client is connected to the Internet via tethering with Nexus7 and BM-FRML-1GBM. The other PC for the ntpd server is connected to the Internet via a 1000BASE-T Ethernet cable. In the mesh network measurement, the configuration of the connection is the same as that used to measure the throughput and round-trip time. Table 6 shows the offset and jitter of NTP on each network obtained by the ntpq command after synchronization for 30 min. According to the results, ssyncerr for WiFi is of 10 ms order, ssyncerr for LTE is of 100 ms order, and ssyncerr for the mesh network is of ms order. Note that the throughput and round-trip time measurements of LTE are not conducted between PCs and the NTP measurements are conducted under the condition that only the client side PC utilizes LTE. This is because of a limitation of the device that the PC is not allowed to obtain a global IP address using LTE. However, the direct connection between PCs can be achieved by using a particular plan provided by a carrier or VPN even for LTE. Therefore, LTE can be used in the proposed method. Table 2 Throughput of each network (Mbps). Average WiFi LTE Mesh network(direct) Mesh network(1 hop)
27.11 1.962 13.50 5.070
/ (deg.)
k (deg.)
Nexus7
1:0289 10
GPS18x LVC
8:2375 106
6
3:3536 10
Ng (m) 7
8:8742 106
H (m)
0
0.60287
5:5390 1012
7.7818
Table 5 Standard deviations of coordinate values calculated with data obtained from GPS (m).
Nexus7 GPS18x LVC
x
y
z
0.39127 4.0248
0.41518 5.5292
0.22824 3.9141
4.3. Signal to interference ratio with/without proposed method In this section, the performance of the proposed method is evaluated in terms of the signal to interference ratio. The signal to interference ratio of a public address system with and without the proposed method is estimated as follows. In this estimation, N nodes are assumed to form a regular polygon with overlapping service areas. The example of N ¼ 3 is shown in Fig. 8. Listening points are located in the service areas of n ðn 6 NÞ nodes, and listening points in overlapping service areas are equidistant from the neighboring nodes. This arrangement of nodes is a rare case, however, it is employed to simulate worse-case scenario. The signal to interference ratio is estimated using Eq. (7) by considering the sound from one node as the signal and the sounds from other nodes as interference. In Fig. 8, for example, the numbers of sounds acting as interference are 0 at listening point A, 1 at B, and 2 at C. The gain factor of each node is assumed to be 1. The impulse responses between nodes and listening points are assumed to have attenuation and delay factors that depend only on the distance. Delay factors mentioned in Section 4.1 other than the delays of D/A conversion do not produce sound overlap in the proposed method. The amount of sound overlap resulted from the delay of D/A conversion is negligible small. On the other hand, error factors measurements in Section 4.2 have the possibility to produce sound overlap as shown in Fig. 9. According to the results of error factors measurements in Section 4.2, the maximum amount of sound overlap resulted from error factors can be considered to be below 200 ms in the proposed method. The duration of the input signal is set to 3 s in this estimation. The absolute value of the amplitude of the input signal is consistently set to 1 over its duration to simulate severe condition. The ratio of the period with SIR = 0 dB to the emission period is calculated with 200 ms of the amount of sound overlap resulted from error factors. Fig. 10 shows the average of this ratio over each node. Without the proposed method, each node emits sound simultaneously, therefore the ratio of the period with SIR = 0 dB to the emission period is consistently 100%. With the proposed method, the amount of sound overlap resulted from error factors increases with increasing the number of interference nodes. However, the ratio with the proposed method is consistently less than the ratio without proposed method. Moreover, because the setting of this estimation is worse-case scenario, the actual impact of error factors to SIR can be considered smaller.
Table 3 Round-trip time of each network (ms). Standard deviation 0.8439 0.5010 0.9920 0.4770
WiFi LTE Mesh network(direct) Mesh network(1 hop)
Average
Standard deviation
58.71 237.4 2.170 3.684
30.93 162.4 0.130 0.409
77
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78 Table 6 Offset and jitter of NTP on each network (ms).
WiFi LTE Mesh network (direct) Mesh network (1 hop)
S3
Offset
Jitter
1.573 220.19 0.558 0.951
28.846 256.934 0.063 0.122
Sound overlap resulted from error factors
S2 SIR2 = +∞
SIR2 = 0
SIR2 = 0
S1
t
5. Demonstration on laptop PC
Emission timing difference resulted from error factors Fig. 9. Example of sound overlap due to error factors.
Raito of period when SIR = 0 dB
100
80
60
40
20
0
0
1
2
3
4
5
Number of interference nodes
Fig. 10. The average ratio of period when SIR = 0 dB of a public address system with/without proposed method.
s3
Index of node
The proposed method is implemented on a laptop PC (Apple Inc. MacBook Pro 2012 Mid). Three nodes, s1 ; s2 , and s3 , are separately launched on three PCs connected to each other via WiFi using the devices listed in Table 1. A server to store the information on the nodes in one area is also launched on another PC. The internal times of the three PCs are preliminarily synchronized using ntpd (ver. 4.2.7p444) instead of a GPS to simulate the worse-case scenario. s2 is set to be the NTP server and s1 and s3 are set to be the NTP client to obtain time information on s2 . The offset and jitter of s1 and s3 are 11.893 ms and 23.816 ms, and 2.859 ms and 33.509 ms, respectively. The coordinate values of the three nodes, (x1 ; y1 ; z1 ), (x2 ; y2 ; z2 ), and (x3 ; y3 ; z3 ), are determined so that the three nodes form a regular triangle such as the one shown in Fig. 8, and the distance between each node is 400 m. Then, the standard deviations of the coordinate values, calculated using the location information obtained by Nexus7 shown in Table 5, are added to the coordinate values of each node. t GPS is set to 10 ms in accordance with the estimation in Section 4.2.1 and tsync is set to 50 ms in accordance with the estimation in Section 4.2.2. The diameter of the service area of each node R is set to 300 m, the height of the loudspeakers H is set to 15 m, the height of the listening points h is set to 1.5 m, and the speed of sound c is set to 340 m/s. The priority value of each node is set so that p1 < p2 < p3 ; therefore, the nodes are expected to emit sound in the order s1 ; s2 ; s3 . After each node finishes the construction of the database containing information on the other nodes, the input signal is fed into s1 . The input signal is a pink noise having 3 s duration with 1 s silence added before and after its duration. The sampling frequency of the input signal is 44100 Hz and the number of quantization bits is 16. In this implementation, 10 frames are used to detect the start of the sound event and three frames are used to detect the end of the sound event. Fig. 11 shows the emission timing of each node. According to Eq. (21), t12 ¼ t23 ¼ 0:586 s. In this trial, the time between the end of emission of s1 and the start of emission of s2 is 0.789 s and the time between the end of emission
s2
s1
0
2
4
6
8
10
12
Time (s) Fig. 11. Emission timing of demonstration system.
of s2 and the start of emission of s3 is 0.859 s. Both these elapsed times exceed t 12 and t 23 . Therefore, the sounds emitted from each node do not overlap at listening points within the service areas of these nodes under the assumptions that the attenuation and delay factor between a loudspeaker and a listening point depends only on the distance and the directivity of each node is omnidirection. 6. Conclusions
Fig. 8. Nodes with overlapping service areas and listening points.
In this paper, to minimize the sound overlap of a public address system for emergency announcements, a method for controlling
78
T. Onoguchi et al. / Applied Acoustics 98 (2015) 70–78
the timing of emission was proposed. The proposed method focuses on the input and output of a public address system. Under the assumptions that the attenuation and delay factor between a loudspeaker and a listening point depends only on the distance and the directivity of each node is assumed to be omnidirection, the optimal emission timing for a pair of nodes is expressed in terms of the signal to interference ratio. Then, on the basis of the above optimal emission timing, the emission timing for a set of nodes in a management area is determined. In the proposed method, each node collects and shares the positions of each node obtained by GPS via a computer network to determine the emission timing, and then taking into account whether other nodes emit sound or not each node autonomously determines its emission timing and the delay time for other nodes. The delay and error factors included in a public address system employing the proposed method are also modeled and estimated. The proposed method can be implemented without any dedicated hardware, as shown in Section 5, and, therefore, can be applied to existing loudspeakers. In future work, more detailed consideration of the signal transmission between a loudspeaker and a listening point and the directivity of node is needed. In particular, the service area of each node is assumed to be a perfect circle in the proposed method. Therefore, a method of controlling the emission level to compensate for the distortion of the service area caused by wind, rainfall, and reflection from buildings should be developed. Acknowledgments Part of this work was carried out with the support of a Grant-in-Aid for the Strategic Information and Communications R&D Promotion Programme (SCOPE) of the Ministry of Internal Affairs and Communications of Japan No. 132310012 ‘‘Intelligent system for public address in the open air with mesh network for transmitting disaster mitigation information’’. References [1] NTT DOCOMO, INC. Area mail disaster information service,
[3] Material of the Committee for Technical Investigation on Countermeasures for Earthquakes and Tsunamis Based on the Lessons Learned from the ‘‘2011 off the Pacific coast of Tohoku Earthquake,’’