- Email: [email protected]
Experimental demonstration of phase-coherent underwater acoustic communications using a compact array
Ocean Engineering 145 (2017) 207–214
Contents lists available at ScienceDirect
Ocean Engineering journal homepage: www.elsevier.com/locate/oceaneng
Experimental demonstration of phase-coherent underwater acoustic communications using a compact array Chengbing He a, *, Qunfei Zhang a, Zhenhua Yan b, Qinyuan Li a, Lingling Zhang a, Jianfeng Chen a, Qian Qi a a b
School of Marine Science and Technology, Northwestern Polytechnical University, Shaanxi, 710072, PR China China Ship Research and Development Academy, Beijing, 100085, PR China
A R T I C L E I N F O
A B S T R A C T
Keywords: Underwater acoustic communication Compact array with closely spaced elements Spatial diversity Bidirectional equalizer Lake experiment
High-rate phase coherent acoustic communication systems generally require large arrays to obtain spatial diversity. Large arrays may not be suitable for small underwater platforms such as underwater unmanned vehicles (UUVs). In this paper, acoustic communication utilizing a compact array with closely spaced elements was demonstrated using the data collected during the August 2010 lake experiment. In addition, a bidirectional multichannel decision feedback equalizer (DFE) was developed to enhance the performance of communication systems. The recorded data from three communication ranges of 25 m, 920 m, and 1 300 m were processed to verify the system performance. The results showed an improvement in output symbol signal-to-noise ratio (OSNR) of approximately 3.4 dB, and 4.9 dB for bidirectional two-channel DFE and four-channel DFE over a singlechannel DFE, respectively.
1. Introduction With the increasing demand for ocean exploring, environmental monitoring, and scientific data collection, reliable and high-rate underwater acoustic (UWA) communication techniques for underwater devices will play an important role in the future. The challenge of high-rate UWA communication lies primarily on the unique characteristics of UWA channels, such as severe attenuation, limited bandwidth, time-varying long spread multipath propagation, channel fading, strong background noise, and the Doppler effect (Kilfoyle and Baggeroer, 2000). Phase coherent single carrier adaptive time-domain decision feedback equalizers (SC-TDE) with an embedded second-order digital phase locked loop (DPLL) have been investigated for decades. The receiver can simultaneously track carrier phases as well as mitigate multipath channel distortions (Stojanovic et al., 1994). To compensate for the effect of channel fading, multichannel receivers are widely used in high-rate phase coherent UWA communication systems (Stojanovic, 2008; Pajovic and Preisig, 2015). The benefit of spatial diversity in improving the output symbol signalto-noise ratio (OSNR) was demonstrated in many experiments (Yang, 2007; Zhang and Dong, 2011; Jamshidi and Moezzi, 2015). Theoretical analysis indicates that, to ensure the fading on each element
approximately independent, the optimal element spacing is about the signal coherence length (Pajovic and Preisig, 2015). For vertical arrays, the optimal element spacing is on the order of 3–4 wavelengths, while the coherence length is on the order of 30–60 wavelengths for horizontal arrays (Yang and Heaney, 2012). This requirement restricts application of conventional spatial diversity on small underwater platforms. In the past few years, small size receivers, such as vector sensors, have been considered for phase coherent underwater acoustic communication (Song et al., 2011; Han et al., 2015). In this paper, we proposed phase coherent underwater acoustic communications using a compact array with closely spaced elements to significantly reduce receiver size. The sensor separation of the compact array is about 1/10 wavelength. A compact array with such spaced elements is generally used as a super-gain hydrophone array to improve detection performance of passive sonar. The small sensor separation in compact arrays reduce spatial diversity gain, and a bidirectional multichannel DFE (BiMCE) was proposed for compact arrays to obtains additional diversity. The DFE suffers from error propagation caused by the feedback of incorrect decisions. To mitigate error propagation and improve the performance of a conventional DFE, bidirectional DFE was first introduced in wireless communication (Balakrishnan and Johnson, 2000), and then refined for complementary code keying (CCK) UWA
* Corresponding author. E-mail address: [email protected] (C. He). http://dx.doi.org/10.1016/j.oceaneng.2017.08.058 Received 16 December 2015; Received in revised form 14 July 2017; Accepted 24 August 2017 0029-8018/© 2017 Elsevier Ltd. All rights reserved.
C. He et al.
Ocean Engineering 145 (2017) 207–214 Table 1 Receiver parameters.
Fig. 1. Block diagram of multichannel decision feedback equalizer with DPLL.
Parameters
Description
Values
fs B Tc M K Np Nf Nb Kf1 Kf2 λ Range Rate d
Sampling frequency Bandwidth Symbol duration Total sensor numbers Oversampling rate The training symbol length Feedforward filter order Backward filter order Proportional tracking constants in PLL Integral tracking constants in PLL RLS forgetting factor in DFE Communication range Communication rate sensor separation
20kHz 4 kHz 0.5 ms 8 2 1000 20/100 20/80 0.001 0.001 0.999 20, 920, 1 300 m 2 kbps λ/10
algorithm widely used in UWA communication are reviewed in this section. The transmitted signal is represented in baseband form as
uðtÞ ¼
X
dðnÞgðt nTÞ;
(1)
n
where d(n) are the M-ary phase shift keying (MPSK) modulated data symbols transmitted every T seconds, and g(t) is the transmitter pulse shape filter. The received signal at the input to the kth equalizer branch is modeled as
Fig. 2. Structure of bidirectional multichannel DFE.
yk ðtÞ ¼
X
dðnÞhk ðt nT; tÞejθk ðtÞ þ wk ðtÞ; k ¼ 1;
… K;
(2)
n
where hk (τ,t) is the overall channel response of the k-th channel (including physical channels and transceiver filters), wk(t) is the additive white Gaussian noise, and θk is the phase rotation caused by symbol timing offset and Doppler shift. Without loss of generality, we assume a sampling rate of 2/T for the received signal. The MCE is effective for removing the ISI induced by multipath propagation. When the channel is unknown, the equalizer tap-weights are determined by minimizing the mean square error (MSE) of the input received data symbols and the recovered data symbols. Then, channel tracking is accomplished by the adaptive algorithm combined recursive least squares (RLS) and a second-order DPLL (Stojanovic et al., 1994). The structure of the MCE is shown in Fig. 1. It consists of a bank of adaptive feedforward filters, one per receiving sensor, followed by a decision feedback filter. The feedforward filter coefficients ak arranges in vector ak, while the feedback filter coefficients b arranges in vector b. The coefficients of the forward and feedback filters are iteratively updated and phase compensation is required separately for each sensor. The order of feedforward and feedback filters are M and N, respectively. The input received signal for the feedforward filter is
Fig. 3. Structure of the horizontal compact array with closely spaced elements.
Fig. 4. Transmitted signal structure.
communication in our previous work (He et al., 2009). It was also adopted for time-reversal underwater acoustic communication (Song, 2012) and single carriers with frequency domain equalizers (He et al., 2012). The proposed BiMCE consists of two parallel MCE structures, one to equalize the received multichannel signal in a causal fashion and the other equalize the time-reversed version of the received multichannel signal in a non-causal fashion. The reminder of this paper is organized as follows. Multichannel DFE with the DPLL algorithm is reviewed in Section 2. Section 3 discusses the proposed method. Section 4 illustrates the experimental configuration, analyzes the recorded data, and presents the experiment results. Conclusions are given in Section 5.
yk ðnÞ ¼ ½yk ðnÞ
yk ðn 1Þ
; …;
yk ðn M þ 1ÞT ;
(3)
where½⋅T denotes the matrix transpose. The MCE embedded with the DPLL algorithm is described as follows: Step 1: The feedforward filtering on the sequence of M input symbols, and phase compensation are performed once per symbol, yielding:
2. Adaptive multichannel DFE
pðnÞ ¼
In conventional SC-TDE communication, transmitted signals are corrupted by multipath and noise interference. To combat channel fading, spatial diversity is achieved by multiple spatially separated sensors. The structure of the multichannel DFE (MCE) and the adaptive
where
X
pk ðnÞ;
b pk ðnÞ ¼ a0k yk ðnÞej θ k ðnÞ ;
208
(4)
(5)
C. He et al.
Ocean Engineering 145 (2017) 207–214
Fig. 5. Time-varying channel impulse response (25 m).
Fig. 6. Time-varying channel impulse response (920 m).
Fig. 7. Time-varying channel impulse response (1300 m).
and to produce an estimate of the intersymbol interference (ISI) caused by these pervious symbols,
and ð⋅Þ0 denotes conjugate transpose. Step 2: The feedback filter operates on the sequence of N decided symbols
b dðnÞ ¼ ½dðn 1Þ;
… dðn NÞ : T
b qðnÞ ¼ b0 dðnÞ:
(6)
Step 3: Calculate the symbol at the input of the decision device,
209
(7)
C. He et al.
Ocean Engineering 145 (2017) 207–214
The coefficients of feedforward filters and feedback filters are updated by MSE¼E{je(n)j2} through the recursive least square (RLS) algorithms. Step 5: The second-order DPLL algorithm equation is
Table 2 Root-mean-square delay spread. Range/m
25 920 1300
RMS/ms Ch.1
Ch.2
Ch.3
Ch.4
Ch.5
Ch.6
Ch.7
Ch.8
0.7 15.5 16.3
2.1 16.8 17.0
2.1 15.9 17.0
0.7 15.3 17.1
0.7 14.7 17.2
0.7 15.1 17.2
0.8 15.9 17.2
0.9 16.4 17.6
b θ n þ Kf1 Φn þ Kf2 θ nþ1 ¼ b
n X
Φi :
(10)
o n Φn ¼ Im pðnÞðdðnÞ þ qðnÞÞ* :
(11)
i¼0
where,
is the output of the equivalent phase detector, Kf1 and Kf2 are the DPLL tracking constants. 3. Proposed structure and algorithms Optimal MCE configuring includes sensor numbers, filter length, and sensor separation. Previous research has shown that sensor separation is an important factor in determining equalizer performance (Pajovic and Preisig, 2015). Optimal sensor separation is usually several times that of the minimum wavelength of the wideband underwater acoustic communication signal. However, sensor separation in the MCE decides the receiver size, which may be impossible for a practical small underwater platform. In this paper, we considered a compact array with closely spaced elements for the multichannel DFE, where sensor separation was about 1/ 10 wavelength of the carrier frequency. This sensor separation was much less than optimal, and hence significantly reduced receiver size. However, the channel and noise correlation between sensors still affected system performance. To reduce receiver size, but maintain performance, we propose a BiMCE for the compact array UWA communication system. The block diagram of the proposed BiMCE receiver structure is displayed in Fig. 2. The proposed method performs two equalizations; a forward MCE on the received signal (upper), and a backward MCE on its time-reversal version. The upper block illustrates a MCE where af,k and bf are k-channel feedforward filter and feedback filter coefficients, respectively. The lower block processes the time-reversal multichannel received signal with a backward MCE where ar,k and br are the corresponding kchannel feedforward filter and feedback filter coefficients, respectively. Furthermore, there is a time-reversal operation before and after the backward MCE, as shown in Fig. 2. The details of MCE are shown in Fig. 1, including the DPLLs. The basic assumption is that error propagation in multichannel DFE is causal, and therefore errors caused by the forward multichannel DFE and the time-reversal multichannel DFE have low correlation. This low correlation characteristic leads to inherent diversity gain (He et al., 2009). The receiver combines the soft output of the forward multichannel DFE e df ðnÞ and the time-reversal multichannel DFE output e dr ðnÞ.
Fig. 8. Channel correlation measured from recorded data. Reference sensor is 4.
8 > d r ðnÞ ¼ dðnÞ þ μ2 ðnÞ :e
Fig. 9. Noise correlation measured from recorded data.
e dðnÞ ¼ pðnÞ qðnÞ X b b ¼ a0k yk ðnÞej θ k ðnÞ b0 dðnÞ:
(12)
where μ1(n) and μ2(n) are estimated error noise. The diversity combining depicted in Fig. 2 can be shown as a weighted linear combination of the two sequences. When the MSE, Eje d½n d½nj2 , for the two data streams are equal, an equal-gain combing is optimal (Balakrishnan and Johnson, 2000). Thus, the soft output of the diversity combining block is written as
(8)
Step 4: The error between the decision and the true value is expressed as
eðnÞ ¼ dðnÞ e dðnÞ;
;
e d r ðnÞ dðnÞ ¼ αe d f ðnÞ þ ð1 αÞe
(9)
where the known training sequence is d(n) when the system is in training mode, and then replaced by b d n when the system is in decision feedback mode. The output MSE is E{je(n)j2}.
¼ dðnÞ þ
μ1 ðnÞ þ μ2 ðnÞ ; 2
where α ¼ 1∕2. 210
(13)
C. He et al.
Ocean Engineering 145 (2017) 207–214
Table 3 Average BERs and OSNRs for BPSK communication experiment. Number of sensors used
Method
25 m
920 m
BER Single Sensor
MCE BiMCE MCE BiMCE MCE BiMCE
Two Sensors Four Sensors
OSNR (dB) 7
8.7 10 0 0 0 0 0
1300 m
BER 2.8 7.1 2.6 0 9.3 0
14.1 15.1 15.6 16.8 16.7 18.6
OSNR (dB) 3
10 104 104 105
BER
OSNR (dB) 3
9.8 10 3.4 103 8.8 104 1 104 2.6 104 0
8.4 10.0 9.9 11.7 10.8 13.1
7.1 8.5 8.8 10.5 10.0 12.0
closely spaced elements was deployed 6 m underwater from the receiving ship. The compact array with closely spaced elements consisted of 8 elements, as shown in Fig. 3. Sensors were 20 mm 30 mm, and sensor center separation was 25 mm, which was about 1/10 wavelength of the 6 kHz center frequency. The array aperture was 195 mm. The receiving ship was anchored during the communication experiment. The transmitting ship was positioned at three different sites, A, B, and C, 25 m, 920 m, and 1 300 m away from the receiving ship, respectively. The transmitting ship drifted freely at each site. During the experiment, the structured signal shown in Fig. 4 was transmitted. The probe signal was 0.1 s linear frequency modulated (LFM) signal with a bandwidth of 4 kHz, followed by a 0.1 s guard period. The communication data was a binary phase shift keying (BPSK) modulated signal with a 6 kHz carrier frequency. The symbol duration for the BPSK signal is 0.5 ms with 2 k/s symbol rate. The communication data signal includes 15424 BPSK symbols, and 1000/2000 symbols were used for training sequence. The sampling frequency of the received signal recorder was 20 kHz. Furthermore, fractional spaced equalizers (two samples per symbol) were used for feedforward filters. The detail parameters of the communication system are given in Table 1.
Fig. 10. Output SNR according to the number of sensors.
4.2. Channel characteristics The channel impulse responses (CIR) estimated by the LFM synchronization signal are shown in Figs. 5–7. The left parts of these figures show the CIR of sensor 1 as a function of time at different distances. The lateral axis is the channel multipath delay spread (ms), and the vertical axis is the measurement time interval (s). The right parts of these figures show the estimated CIRs of channels 1, 3, 5, and 7. The left parts of three figures reflect the channel temporal varying, while the right parts of three figures reflect the channel spatial varying. Furthermore, the multipath delay spread was about 8 ms at the range of 25 m, which was equivalent to 16 symbols duration. Conversely, when the communication distance was 920 and 1 300 m, the maximum multipath delay spread was about 50 ms, equivalent to 100 symbols duration. Based on the estimated CIRs in Figs. 5–7, the root-mean-square (RMS) delay spread at different communication distances was calculated according to the following equation:
τrms
Fig. 11. Output SNR according to different sensor separation for two-channel equalizer.
4. Lake experiment results
ffi vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi uP u ðlT Þ2 hðlÞj2 0P ðlT ÞhðlÞj2 12 u s s u B C ¼u l @ l A t P P 2 hðlÞj2 hðlÞj l
(14)
l
where h(l) is a discrete, baseband impulse response sampled at period Ts. The estimated RMS is shown in Table 2. A small delay spread corresponded to less ISI, and shorter distances might offer better communication results than longer distances according to the estimated CIR (Song et al., 2011). The received SNRs were 25 dB, 15.7 dB and 17.1 dB for 25 m, 920 m, and 1 300 m, respectively. Due to the use of a horizontal compact array with closely spaced elements, the correlation was high between different channels at the same time. This correlation between different sensors was analyzed according to the following equation:
4.1. Experimental setup To demonstrate the performance of phase coherent UWA communication using a compact array with closely spaced elements and analyze performance, a communication lake experiment was conducted at Fuxian Lake, Yunnan province, China, in August 2010. The depth of the lake experimental area was approximately 60 m, and a flextensional acoustic transducer was suspended 22 m underwater from the transmitting ship. Additionally, a compact horizontal array with
211
C. He et al.
Ocean Engineering 145 (2017) 207–214
Fig. 12. Single-channel DFE output signal constellation and DPLL outputs (Sensor 3).
Fig. 13. Two-channel DFE output signal constellation and DPLL outputs (Sensor 3 and 4).
γ m;n
E um u*n E½um E u*n ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 2 2 E um j E½um j E un j E½un j
:
as shown in Fig. 9.
(15)
4.3. Results and analysis For convention, the underwater communication performance is often evaluated by the output symbol signal-to-noise ratio. This OSNR is defined as
Fig. 8 presents the channel correlation properties between different sensors utilizing the fourth sensor as a reference. The correlation was high between different elements, because of small sensor separation. As separation increased, correlation became smaller. The noise correlation calculation used 3 s background noise recorded before the probe signal,
SNRoutput ¼
212
1 : MSE
(16)
C. He et al.
Ocean Engineering 145 (2017) 207–214
Fig. 14. Four-channel DFE output signal constellation and DPLL outputs (Sensor 1,2,3,4).
bidirectional four-channel DFE had the average ONSRs improvement of 3.3 dB and 4.7 dB improvement, respectively. Lastly, for the 1300 m range, the single-channel DFE had an average OSNR of 7.1 dB, while the two-channel DFE and the four-channel DFE had the average OSNRs improvements of 1.7 dB and 2.9 dB. The bidirectional two-channel DFE and bidirectional four-channel DFE had average OSNRs improvement of 3.4 dB and 4.9 dB, respectively. Fig. 10 shows the results of the measured OSNR as a function of the element number. As the element number increased from 1 to 4, OSNR improvement was considerable. However, further increasing the element number with the same separation, resulted in only slight OSNR improvement and a huge increase in the receiver complexity. Fig. 11 shows the results of the measured OSNR as a function of the 1 element spacing d ¼ 10 λ ¼ 25 mm for two-channel DFE. With increasing element spacing within this small fixed aperture, OSNR improvement of two-channel MCE and BiMCE was slightly at the expense of increasing the size of receivers. Further more, OSNR improvement in the twochannel BiMCE over two-channel MCE is almost fixed when the sensor spacing increase. For the range of 920 m and 1300 m, the average OSNR for two-channel BiMCE over two-channel MCE was 1.93 dB and 1.80 dB, respectively. For the range of 25 m, the average OSNR for two-channel BiMCE over two-channel MCE was 1.39 dB. The reason of less average OSNR improvement at 25 m was due to fewer taps of feedforward and feedback filters. Figs. 12–14 show the receiver performance of a single-channel DFE (sensor 4), two-channel DFE (sensor 3,4) and four-channel DFE (sensor 1,2,3, and 4) at 920 m communication distance. The left parts of these figures show the output symbol constellation diagrams of MCE and BiMCE along with the DPLL outputs. For all MCE DPLL outputs, the slopes of the DPLL outputs curve were about 0.97 rad/s, which is equivalent to a frequency shift of 0.15 Hz. For the BiMCE DPLL outputs, half of the curve slopes were about 0.97 rad/s, which was caused by the time-reversal operation in the backward MCE. The number of sensors was increased from 1 to 4, the error bits decreased from 16 to 0, and the OSNR increased from 8.6 dB to 13 dB. Consequently, OSNR improvements for BiMCE were affected by the number of sensors and feedforward and feedback filter taps, and also the channel conditions.
Table 3 illustrates the average OSNR and bit error rate (BER) of a single-channel DFE, two-channel DFE and four-channel DFE and their corresponding bidirectional multichannel DFEs. Considering the complexity of the receiver, a maximum of four channels was used for data processing. Sensors used for these DFEs were chosen consecutively from the 8-element array. For example, the two-channel DFE results were the average results of sensor groups (1,2), (3,4), (5,6), and (7,8), while the four-channel DFE results were the average results of sensor groups (1,2,3,4), (2,3,4,5), (3,4,5,6), and (5,6,7,8). The demodulation results of BPSK transmissions at three ranges are shown in Table 3. At short range (25 m), the number of feedforward taps for each sensor was KNf ¼ 2 20 ¼ 40 for the fractionally spaced DFE, where Nf ¼ 20 was the feedforward filter span in symbols according to the estimated channel in Fig. 5. The number of feedback taps was 20. The total number of taps for the single-channel, two-channel, and fourchannel was 60, 100, and 180, respectively. The training sequence length was 1000. Both the two-channel DFE and the four-channel DFE outperform the single-channel DFE (average OSNR ¼ 14.1 dB) with average OSNRs 1.5 dB and 2.6 dB, respectively. Furthermore, the bidirectional DFE further improved the average OSNR. The bidirectional two-channel DFE and bidirectional four-channel DFE had the average ONSRs 2.7 dB and 4.5 dB improvement, respectively. Although, the performance of the bidirectional two-channel DFE was slightly better than the four-channel DFE, the receiver size was reduced by in the twochannel DFE. At the middle range (920 m, 1300 m), the number of feedforward taps for each sensor was KNf ¼ 2 100 ¼ 200 for the fractionally spaced DFE, where Nf ¼ 100 was the feedforward filter span in symbols according to the estimated channel in Figs. 6–7. The number of feedback taps was 80. The total number of the taps for single-channel, two-channel and fourchannel was 280, 480, and 880 respectively. The training sequence length was 1000 for the single-channel and two-channel DFEs, and 2000 for the four-channel DFE. For the 920 m range, the single-channel DFE (average OSNR ¼ 8.4 dB), was outperformed by both the two-channel DFE (average OSNR 1.5 dB improvement) and the four-channel DFE (average OSNR 2.4 dB improvement). Again, the bidirectional DFE further improve the average OSNR. The bidirectional two-channel DFE and 213
C. He et al.
Ocean Engineering 145 (2017) 207–214
5. Conclusion
References
In this paper, we utilized a compact horizontal array with closely spaced elements with a 1/10 wavelength sensor separation for a phase coherent high-rate underwater acoustic communication systems. Through the use of our lake experimental data, coherent communication by the compact array with closely spaced elements was demonstrated at three different communication ranges. To compensate for performance loss caused by small sensor separation, a bidirectional multichannel DFE was proposed for the closely spaced multichannel DFE. Two kinds of equalizers were implemented to process the data. Channel parameters such as channel/noise correlation, and RMS delay spreads were calculated from the measurements. Results showed that the bidirectional multichannel DFE using a compact horizontal array with closely spaced elements offered significant reduction in receiver size for coherent acoustic communication at the carrier frequency of 6 kHz. Furthermore, both MCE and BiMCE performed significantly better than a single sensor. Our results suggest that the compact horizontal array with the proposed BiMCE is effective for phase coherent underwater acoustic communications systems particularly required for the small underwater platforms.
Balakrishnan, J., Johnson Jr., C.R., Oct. 2000. Time-reversal diversity in decision feedback equalization. In: Proc. of Allerton Conf. on Comm. Control and Computing. Monticello, IL, USA. Han, X., Yin, J., Yu, G., Yu, P., 2015. Experimental demonstration of single carrier underwater acoustic communication using a vector sensor. Appl. Acoust. 98, 1–5. He, C.B., Huang, J.G., Ding, Z., Oct. 2009. A variable-rate spread-spectrum system for underwater acoustic communications. IEEE J. Ocean. Eng. 34 (4), 624–633. He, C.B., Huang, J.G., Zhang, Q.F. Hybrid Time-Frequency Domain Equalization for Single-carrier Underwater Acoustic Communications WUWNet'12, Nov.5–6, 2012, Los Angeles, California, USA. Jamshidi, A., Moezzi, S., 2015. Experiential assessment of iteratively residual interference elimination in the passive phase conjugation for acoustic underwater communications. Ocean. Eng. 105, 287–294. Kilfoyle, D.B., Baggeroer, A.B., Jan. 2000. The state of the art in underwater acoustic telemetry. IEEE J. Ocean. Eng. 25 (1), 4–27. Pajovic, Milutin, Preisig, James C., 2015. Performance analysis and optimal design of multichannel equalizer for underwater acoustic communications. IEEE J. Ocean. Eng. 40 (4), 759–774. Song, H.C., April 2012. Bidirectional equalization for underwater acoustic communication. J. Acoust. Soc. Am. 131 (4), 342–347. Song, Aijun, Abdi, Ali, Badiey, Mohsen, Hursky, Paul, 2011. Experimental demonstration of underwater acoustic communication by vector sensors. IEEE J. Ocean. Eng. 36 (3), 454–461. Stojanovic, M., Jul. 2008. Efficient processing of acoustic signals for high rate information transmission over sparse underwater channels. Elsevier J. Phys. Commun. 146–161. Stojanovic, M., Catipovic, J.A., Proakis, J.G., Jan. 1994. Phase-coherent digital communications for underwater acoustic channels. IEEE J. Ocean. Eng. 19 (1), 100–111. Yang, T.C., 2007. A study of spatial processing gain in underwater acoustic communications. IEEE J. Ocean. Eng. 32 (3), 689–709. Yang, T.C., Kevin D, Heaney, Network-assisted Underwater Acoustic Communications, WUWNet'12, Los Angeles, CA, USA, Nov. 5-6, 2012. Zhang, G.S., Dong, H.F., 2011. Spatial diversity in multichannel processing for underwater acoustic communications. Ocean. Eng. 38 (14), 1611–1623.
Acknowledgments This work was supported in part by the National Key R&D Program of China under Grant 2016YFC1400200, the National Natural Science Foundation of China (Grant No. 61471298), and the Fundamental Research Funds for the Central Universities(Grant No.3102016AXXX03). The authors would like to thank all the participants of the lake experiment. The authors would also like to thank the anonymous reviewers for their valuable comments and suggestions on the original manuscript.
214
Contents lists available at ScienceDirect
Ocean Engineering journal homepage: www.elsevier.com/locate/oceaneng
Experimental demonstration of phase-coherent underwater acoustic communications using a compact array Chengbing He a, *, Qunfei Zhang a, Zhenhua Yan b, Qinyuan Li a, Lingling Zhang a, Jianfeng Chen a, Qian Qi a a b
School of Marine Science and Technology, Northwestern Polytechnical University, Shaanxi, 710072, PR China China Ship Research and Development Academy, Beijing, 100085, PR China
A R T I C L E I N F O
A B S T R A C T
Keywords: Underwater acoustic communication Compact array with closely spaced elements Spatial diversity Bidirectional equalizer Lake experiment
High-rate phase coherent acoustic communication systems generally require large arrays to obtain spatial diversity. Large arrays may not be suitable for small underwater platforms such as underwater unmanned vehicles (UUVs). In this paper, acoustic communication utilizing a compact array with closely spaced elements was demonstrated using the data collected during the August 2010 lake experiment. In addition, a bidirectional multichannel decision feedback equalizer (DFE) was developed to enhance the performance of communication systems. The recorded data from three communication ranges of 25 m, 920 m, and 1 300 m were processed to verify the system performance. The results showed an improvement in output symbol signal-to-noise ratio (OSNR) of approximately 3.4 dB, and 4.9 dB for bidirectional two-channel DFE and four-channel DFE over a singlechannel DFE, respectively.
1. Introduction With the increasing demand for ocean exploring, environmental monitoring, and scientific data collection, reliable and high-rate underwater acoustic (UWA) communication techniques for underwater devices will play an important role in the future. The challenge of high-rate UWA communication lies primarily on the unique characteristics of UWA channels, such as severe attenuation, limited bandwidth, time-varying long spread multipath propagation, channel fading, strong background noise, and the Doppler effect (Kilfoyle and Baggeroer, 2000). Phase coherent single carrier adaptive time-domain decision feedback equalizers (SC-TDE) with an embedded second-order digital phase locked loop (DPLL) have been investigated for decades. The receiver can simultaneously track carrier phases as well as mitigate multipath channel distortions (Stojanovic et al., 1994). To compensate for the effect of channel fading, multichannel receivers are widely used in high-rate phase coherent UWA communication systems (Stojanovic, 2008; Pajovic and Preisig, 2015). The benefit of spatial diversity in improving the output symbol signalto-noise ratio (OSNR) was demonstrated in many experiments (Yang, 2007; Zhang and Dong, 2011; Jamshidi and Moezzi, 2015). Theoretical analysis indicates that, to ensure the fading on each element
approximately independent, the optimal element spacing is about the signal coherence length (Pajovic and Preisig, 2015). For vertical arrays, the optimal element spacing is on the order of 3–4 wavelengths, while the coherence length is on the order of 30–60 wavelengths for horizontal arrays (Yang and Heaney, 2012). This requirement restricts application of conventional spatial diversity on small underwater platforms. In the past few years, small size receivers, such as vector sensors, have been considered for phase coherent underwater acoustic communication (Song et al., 2011; Han et al., 2015). In this paper, we proposed phase coherent underwater acoustic communications using a compact array with closely spaced elements to significantly reduce receiver size. The sensor separation of the compact array is about 1/10 wavelength. A compact array with such spaced elements is generally used as a super-gain hydrophone array to improve detection performance of passive sonar. The small sensor separation in compact arrays reduce spatial diversity gain, and a bidirectional multichannel DFE (BiMCE) was proposed for compact arrays to obtains additional diversity. The DFE suffers from error propagation caused by the feedback of incorrect decisions. To mitigate error propagation and improve the performance of a conventional DFE, bidirectional DFE was first introduced in wireless communication (Balakrishnan and Johnson, 2000), and then refined for complementary code keying (CCK) UWA
* Corresponding author. E-mail address: [email protected] (C. He). http://dx.doi.org/10.1016/j.oceaneng.2017.08.058 Received 16 December 2015; Received in revised form 14 July 2017; Accepted 24 August 2017 0029-8018/© 2017 Elsevier Ltd. All rights reserved.
C. He et al.
Ocean Engineering 145 (2017) 207–214 Table 1 Receiver parameters.
Fig. 1. Block diagram of multichannel decision feedback equalizer with DPLL.
Parameters
Description
Values
fs B Tc M K Np Nf Nb Kf1 Kf2 λ Range Rate d
Sampling frequency Bandwidth Symbol duration Total sensor numbers Oversampling rate The training symbol length Feedforward filter order Backward filter order Proportional tracking constants in PLL Integral tracking constants in PLL RLS forgetting factor in DFE Communication range Communication rate sensor separation
20kHz 4 kHz 0.5 ms 8 2 1000 20/100 20/80 0.001 0.001 0.999 20, 920, 1 300 m 2 kbps λ/10
algorithm widely used in UWA communication are reviewed in this section. The transmitted signal is represented in baseband form as
uðtÞ ¼
X
dðnÞgðt nTÞ;
(1)
n
where d(n) are the M-ary phase shift keying (MPSK) modulated data symbols transmitted every T seconds, and g(t) is the transmitter pulse shape filter. The received signal at the input to the kth equalizer branch is modeled as
Fig. 2. Structure of bidirectional multichannel DFE.
yk ðtÞ ¼
X
dðnÞhk ðt nT; tÞejθk ðtÞ þ wk ðtÞ; k ¼ 1;
… K;
(2)
n
where hk (τ,t) is the overall channel response of the k-th channel (including physical channels and transceiver filters), wk(t) is the additive white Gaussian noise, and θk is the phase rotation caused by symbol timing offset and Doppler shift. Without loss of generality, we assume a sampling rate of 2/T for the received signal. The MCE is effective for removing the ISI induced by multipath propagation. When the channel is unknown, the equalizer tap-weights are determined by minimizing the mean square error (MSE) of the input received data symbols and the recovered data symbols. Then, channel tracking is accomplished by the adaptive algorithm combined recursive least squares (RLS) and a second-order DPLL (Stojanovic et al., 1994). The structure of the MCE is shown in Fig. 1. It consists of a bank of adaptive feedforward filters, one per receiving sensor, followed by a decision feedback filter. The feedforward filter coefficients ak arranges in vector ak, while the feedback filter coefficients b arranges in vector b. The coefficients of the forward and feedback filters are iteratively updated and phase compensation is required separately for each sensor. The order of feedforward and feedback filters are M and N, respectively. The input received signal for the feedforward filter is
Fig. 3. Structure of the horizontal compact array with closely spaced elements.
Fig. 4. Transmitted signal structure.
communication in our previous work (He et al., 2009). It was also adopted for time-reversal underwater acoustic communication (Song, 2012) and single carriers with frequency domain equalizers (He et al., 2012). The proposed BiMCE consists of two parallel MCE structures, one to equalize the received multichannel signal in a causal fashion and the other equalize the time-reversed version of the received multichannel signal in a non-causal fashion. The reminder of this paper is organized as follows. Multichannel DFE with the DPLL algorithm is reviewed in Section 2. Section 3 discusses the proposed method. Section 4 illustrates the experimental configuration, analyzes the recorded data, and presents the experiment results. Conclusions are given in Section 5.
yk ðnÞ ¼ ½yk ðnÞ
yk ðn 1Þ
; …;
yk ðn M þ 1ÞT ;
(3)
where½⋅T denotes the matrix transpose. The MCE embedded with the DPLL algorithm is described as follows: Step 1: The feedforward filtering on the sequence of M input symbols, and phase compensation are performed once per symbol, yielding:
2. Adaptive multichannel DFE
pðnÞ ¼
In conventional SC-TDE communication, transmitted signals are corrupted by multipath and noise interference. To combat channel fading, spatial diversity is achieved by multiple spatially separated sensors. The structure of the multichannel DFE (MCE) and the adaptive
where
X
pk ðnÞ;
b pk ðnÞ ¼ a0k yk ðnÞej θ k ðnÞ ;
208
(4)
(5)
C. He et al.
Ocean Engineering 145 (2017) 207–214
Fig. 5. Time-varying channel impulse response (25 m).
Fig. 6. Time-varying channel impulse response (920 m).
Fig. 7. Time-varying channel impulse response (1300 m).
and to produce an estimate of the intersymbol interference (ISI) caused by these pervious symbols,
and ð⋅Þ0 denotes conjugate transpose. Step 2: The feedback filter operates on the sequence of N decided symbols
b dðnÞ ¼ ½dðn 1Þ;
… dðn NÞ : T
b qðnÞ ¼ b0 dðnÞ:
(6)
Step 3: Calculate the symbol at the input of the decision device,
209
(7)
C. He et al.
Ocean Engineering 145 (2017) 207–214
The coefficients of feedforward filters and feedback filters are updated by MSE¼E{je(n)j2} through the recursive least square (RLS) algorithms. Step 5: The second-order DPLL algorithm equation is
Table 2 Root-mean-square delay spread. Range/m
25 920 1300
RMS/ms Ch.1
Ch.2
Ch.3
Ch.4
Ch.5
Ch.6
Ch.7
Ch.8
0.7 15.5 16.3
2.1 16.8 17.0
2.1 15.9 17.0
0.7 15.3 17.1
0.7 14.7 17.2
0.7 15.1 17.2
0.8 15.9 17.2
0.9 16.4 17.6
b θ n þ Kf1 Φn þ Kf2 θ nþ1 ¼ b
n X
Φi :
(10)
o n Φn ¼ Im pðnÞðdðnÞ þ qðnÞÞ* :
(11)
i¼0
where,
is the output of the equivalent phase detector, Kf1 and Kf2 are the DPLL tracking constants. 3. Proposed structure and algorithms Optimal MCE configuring includes sensor numbers, filter length, and sensor separation. Previous research has shown that sensor separation is an important factor in determining equalizer performance (Pajovic and Preisig, 2015). Optimal sensor separation is usually several times that of the minimum wavelength of the wideband underwater acoustic communication signal. However, sensor separation in the MCE decides the receiver size, which may be impossible for a practical small underwater platform. In this paper, we considered a compact array with closely spaced elements for the multichannel DFE, where sensor separation was about 1/ 10 wavelength of the carrier frequency. This sensor separation was much less than optimal, and hence significantly reduced receiver size. However, the channel and noise correlation between sensors still affected system performance. To reduce receiver size, but maintain performance, we propose a BiMCE for the compact array UWA communication system. The block diagram of the proposed BiMCE receiver structure is displayed in Fig. 2. The proposed method performs two equalizations; a forward MCE on the received signal (upper), and a backward MCE on its time-reversal version. The upper block illustrates a MCE where af,k and bf are k-channel feedforward filter and feedback filter coefficients, respectively. The lower block processes the time-reversal multichannel received signal with a backward MCE where ar,k and br are the corresponding kchannel feedforward filter and feedback filter coefficients, respectively. Furthermore, there is a time-reversal operation before and after the backward MCE, as shown in Fig. 2. The details of MCE are shown in Fig. 1, including the DPLLs. The basic assumption is that error propagation in multichannel DFE is causal, and therefore errors caused by the forward multichannel DFE and the time-reversal multichannel DFE have low correlation. This low correlation characteristic leads to inherent diversity gain (He et al., 2009). The receiver combines the soft output of the forward multichannel DFE e df ðnÞ and the time-reversal multichannel DFE output e dr ðnÞ.
Fig. 8. Channel correlation measured from recorded data. Reference sensor is 4.
8 >
Fig. 9. Noise correlation measured from recorded data.
e dðnÞ ¼ pðnÞ qðnÞ X b b ¼ a0k yk ðnÞej θ k ðnÞ b0 dðnÞ:
(12)
where μ1(n) and μ2(n) are estimated error noise. The diversity combining depicted in Fig. 2 can be shown as a weighted linear combination of the two sequences. When the MSE, Eje d½n d½nj2 , for the two data streams are equal, an equal-gain combing is optimal (Balakrishnan and Johnson, 2000). Thus, the soft output of the diversity combining block is written as
(8)
Step 4: The error between the decision and the true value is expressed as
eðnÞ ¼ dðnÞ e dðnÞ;
;
e d r ðnÞ dðnÞ ¼ αe d f ðnÞ þ ð1 αÞe
(9)
where the known training sequence is d(n) when the system is in training mode, and then replaced by b d n when the system is in decision feedback mode. The output MSE is E{je(n)j2}.
¼ dðnÞ þ
μ1 ðnÞ þ μ2 ðnÞ ; 2
where α ¼ 1∕2. 210
(13)
C. He et al.
Ocean Engineering 145 (2017) 207–214
Table 3 Average BERs and OSNRs for BPSK communication experiment. Number of sensors used
Method
25 m
920 m
BER Single Sensor
MCE BiMCE MCE BiMCE MCE BiMCE
Two Sensors Four Sensors
OSNR (dB) 7
8.7 10 0 0 0 0 0
1300 m
BER 2.8 7.1 2.6 0 9.3 0
14.1 15.1 15.6 16.8 16.7 18.6
OSNR (dB) 3
10 104 104 105
BER
OSNR (dB) 3
9.8 10 3.4 103 8.8 104 1 104 2.6 104 0
8.4 10.0 9.9 11.7 10.8 13.1
7.1 8.5 8.8 10.5 10.0 12.0
closely spaced elements was deployed 6 m underwater from the receiving ship. The compact array with closely spaced elements consisted of 8 elements, as shown in Fig. 3. Sensors were 20 mm 30 mm, and sensor center separation was 25 mm, which was about 1/10 wavelength of the 6 kHz center frequency. The array aperture was 195 mm. The receiving ship was anchored during the communication experiment. The transmitting ship was positioned at three different sites, A, B, and C, 25 m, 920 m, and 1 300 m away from the receiving ship, respectively. The transmitting ship drifted freely at each site. During the experiment, the structured signal shown in Fig. 4 was transmitted. The probe signal was 0.1 s linear frequency modulated (LFM) signal with a bandwidth of 4 kHz, followed by a 0.1 s guard period. The communication data was a binary phase shift keying (BPSK) modulated signal with a 6 kHz carrier frequency. The symbol duration for the BPSK signal is 0.5 ms with 2 k/s symbol rate. The communication data signal includes 15424 BPSK symbols, and 1000/2000 symbols were used for training sequence. The sampling frequency of the received signal recorder was 20 kHz. Furthermore, fractional spaced equalizers (two samples per symbol) were used for feedforward filters. The detail parameters of the communication system are given in Table 1.
Fig. 10. Output SNR according to the number of sensors.
4.2. Channel characteristics The channel impulse responses (CIR) estimated by the LFM synchronization signal are shown in Figs. 5–7. The left parts of these figures show the CIR of sensor 1 as a function of time at different distances. The lateral axis is the channel multipath delay spread (ms), and the vertical axis is the measurement time interval (s). The right parts of these figures show the estimated CIRs of channels 1, 3, 5, and 7. The left parts of three figures reflect the channel temporal varying, while the right parts of three figures reflect the channel spatial varying. Furthermore, the multipath delay spread was about 8 ms at the range of 25 m, which was equivalent to 16 symbols duration. Conversely, when the communication distance was 920 and 1 300 m, the maximum multipath delay spread was about 50 ms, equivalent to 100 symbols duration. Based on the estimated CIRs in Figs. 5–7, the root-mean-square (RMS) delay spread at different communication distances was calculated according to the following equation:
τrms
Fig. 11. Output SNR according to different sensor separation for two-channel equalizer.
4. Lake experiment results
ffi vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi uP u ðlT Þ2 hðlÞj2 0P ðlT ÞhðlÞj2 12 u s s u B C ¼u l @ l A t P P 2 hðlÞj2 hðlÞj l
(14)
l
where h(l) is a discrete, baseband impulse response sampled at period Ts. The estimated RMS is shown in Table 2. A small delay spread corresponded to less ISI, and shorter distances might offer better communication results than longer distances according to the estimated CIR (Song et al., 2011). The received SNRs were 25 dB, 15.7 dB and 17.1 dB for 25 m, 920 m, and 1 300 m, respectively. Due to the use of a horizontal compact array with closely spaced elements, the correlation was high between different channels at the same time. This correlation between different sensors was analyzed according to the following equation:
4.1. Experimental setup To demonstrate the performance of phase coherent UWA communication using a compact array with closely spaced elements and analyze performance, a communication lake experiment was conducted at Fuxian Lake, Yunnan province, China, in August 2010. The depth of the lake experimental area was approximately 60 m, and a flextensional acoustic transducer was suspended 22 m underwater from the transmitting ship. Additionally, a compact horizontal array with
211
C. He et al.
Ocean Engineering 145 (2017) 207–214
Fig. 12. Single-channel DFE output signal constellation and DPLL outputs (Sensor 3).
Fig. 13. Two-channel DFE output signal constellation and DPLL outputs (Sensor 3 and 4).
γ m;n
E um u*n E½um E u*n ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 2 2 2 E um j E½um j E un j E½un j
:
as shown in Fig. 9.
(15)
4.3. Results and analysis For convention, the underwater communication performance is often evaluated by the output symbol signal-to-noise ratio. This OSNR is defined as
Fig. 8 presents the channel correlation properties between different sensors utilizing the fourth sensor as a reference. The correlation was high between different elements, because of small sensor separation. As separation increased, correlation became smaller. The noise correlation calculation used 3 s background noise recorded before the probe signal,
SNRoutput ¼
212
1 : MSE
(16)
C. He et al.
Ocean Engineering 145 (2017) 207–214
Fig. 14. Four-channel DFE output signal constellation and DPLL outputs (Sensor 1,2,3,4).
bidirectional four-channel DFE had the average ONSRs improvement of 3.3 dB and 4.7 dB improvement, respectively. Lastly, for the 1300 m range, the single-channel DFE had an average OSNR of 7.1 dB, while the two-channel DFE and the four-channel DFE had the average OSNRs improvements of 1.7 dB and 2.9 dB. The bidirectional two-channel DFE and bidirectional four-channel DFE had average OSNRs improvement of 3.4 dB and 4.9 dB, respectively. Fig. 10 shows the results of the measured OSNR as a function of the element number. As the element number increased from 1 to 4, OSNR improvement was considerable. However, further increasing the element number with the same separation, resulted in only slight OSNR improvement and a huge increase in the receiver complexity. Fig. 11 shows the results of the measured OSNR as a function of the 1 element spacing d ¼ 10 λ ¼ 25 mm for two-channel DFE. With increasing element spacing within this small fixed aperture, OSNR improvement of two-channel MCE and BiMCE was slightly at the expense of increasing the size of receivers. Further more, OSNR improvement in the twochannel BiMCE over two-channel MCE is almost fixed when the sensor spacing increase. For the range of 920 m and 1300 m, the average OSNR for two-channel BiMCE over two-channel MCE was 1.93 dB and 1.80 dB, respectively. For the range of 25 m, the average OSNR for two-channel BiMCE over two-channel MCE was 1.39 dB. The reason of less average OSNR improvement at 25 m was due to fewer taps of feedforward and feedback filters. Figs. 12–14 show the receiver performance of a single-channel DFE (sensor 4), two-channel DFE (sensor 3,4) and four-channel DFE (sensor 1,2,3, and 4) at 920 m communication distance. The left parts of these figures show the output symbol constellation diagrams of MCE and BiMCE along with the DPLL outputs. For all MCE DPLL outputs, the slopes of the DPLL outputs curve were about 0.97 rad/s, which is equivalent to a frequency shift of 0.15 Hz. For the BiMCE DPLL outputs, half of the curve slopes were about 0.97 rad/s, which was caused by the time-reversal operation in the backward MCE. The number of sensors was increased from 1 to 4, the error bits decreased from 16 to 0, and the OSNR increased from 8.6 dB to 13 dB. Consequently, OSNR improvements for BiMCE were affected by the number of sensors and feedforward and feedback filter taps, and also the channel conditions.
Table 3 illustrates the average OSNR and bit error rate (BER) of a single-channel DFE, two-channel DFE and four-channel DFE and their corresponding bidirectional multichannel DFEs. Considering the complexity of the receiver, a maximum of four channels was used for data processing. Sensors used for these DFEs were chosen consecutively from the 8-element array. For example, the two-channel DFE results were the average results of sensor groups (1,2), (3,4), (5,6), and (7,8), while the four-channel DFE results were the average results of sensor groups (1,2,3,4), (2,3,4,5), (3,4,5,6), and (5,6,7,8). The demodulation results of BPSK transmissions at three ranges are shown in Table 3. At short range (25 m), the number of feedforward taps for each sensor was KNf ¼ 2 20 ¼ 40 for the fractionally spaced DFE, where Nf ¼ 20 was the feedforward filter span in symbols according to the estimated channel in Fig. 5. The number of feedback taps was 20. The total number of taps for the single-channel, two-channel, and fourchannel was 60, 100, and 180, respectively. The training sequence length was 1000. Both the two-channel DFE and the four-channel DFE outperform the single-channel DFE (average OSNR ¼ 14.1 dB) with average OSNRs 1.5 dB and 2.6 dB, respectively. Furthermore, the bidirectional DFE further improved the average OSNR. The bidirectional two-channel DFE and bidirectional four-channel DFE had the average ONSRs 2.7 dB and 4.5 dB improvement, respectively. Although, the performance of the bidirectional two-channel DFE was slightly better than the four-channel DFE, the receiver size was reduced by in the twochannel DFE. At the middle range (920 m, 1300 m), the number of feedforward taps for each sensor was KNf ¼ 2 100 ¼ 200 for the fractionally spaced DFE, where Nf ¼ 100 was the feedforward filter span in symbols according to the estimated channel in Figs. 6–7. The number of feedback taps was 80. The total number of the taps for single-channel, two-channel and fourchannel was 280, 480, and 880 respectively. The training sequence length was 1000 for the single-channel and two-channel DFEs, and 2000 for the four-channel DFE. For the 920 m range, the single-channel DFE (average OSNR ¼ 8.4 dB), was outperformed by both the two-channel DFE (average OSNR 1.5 dB improvement) and the four-channel DFE (average OSNR 2.4 dB improvement). Again, the bidirectional DFE further improve the average OSNR. The bidirectional two-channel DFE and 213
C. He et al.
Ocean Engineering 145 (2017) 207–214
5. Conclusion
References
In this paper, we utilized a compact horizontal array with closely spaced elements with a 1/10 wavelength sensor separation for a phase coherent high-rate underwater acoustic communication systems. Through the use of our lake experimental data, coherent communication by the compact array with closely spaced elements was demonstrated at three different communication ranges. To compensate for performance loss caused by small sensor separation, a bidirectional multichannel DFE was proposed for the closely spaced multichannel DFE. Two kinds of equalizers were implemented to process the data. Channel parameters such as channel/noise correlation, and RMS delay spreads were calculated from the measurements. Results showed that the bidirectional multichannel DFE using a compact horizontal array with closely spaced elements offered significant reduction in receiver size for coherent acoustic communication at the carrier frequency of 6 kHz. Furthermore, both MCE and BiMCE performed significantly better than a single sensor. Our results suggest that the compact horizontal array with the proposed BiMCE is effective for phase coherent underwater acoustic communications systems particularly required for the small underwater platforms.
Balakrishnan, J., Johnson Jr., C.R., Oct. 2000. Time-reversal diversity in decision feedback equalization. In: Proc. of Allerton Conf. on Comm. Control and Computing. Monticello, IL, USA. Han, X., Yin, J., Yu, G., Yu, P., 2015. Experimental demonstration of single carrier underwater acoustic communication using a vector sensor. Appl. Acoust. 98, 1–5. He, C.B., Huang, J.G., Ding, Z., Oct. 2009. A variable-rate spread-spectrum system for underwater acoustic communications. IEEE J. Ocean. Eng. 34 (4), 624–633. He, C.B., Huang, J.G., Zhang, Q.F. Hybrid Time-Frequency Domain Equalization for Single-carrier Underwater Acoustic Communications WUWNet'12, Nov.5–6, 2012, Los Angeles, California, USA. Jamshidi, A., Moezzi, S., 2015. Experiential assessment of iteratively residual interference elimination in the passive phase conjugation for acoustic underwater communications. Ocean. Eng. 105, 287–294. Kilfoyle, D.B., Baggeroer, A.B., Jan. 2000. The state of the art in underwater acoustic telemetry. IEEE J. Ocean. Eng. 25 (1), 4–27. Pajovic, Milutin, Preisig, James C., 2015. Performance analysis and optimal design of multichannel equalizer for underwater acoustic communications. IEEE J. Ocean. Eng. 40 (4), 759–774. Song, H.C., April 2012. Bidirectional equalization for underwater acoustic communication. J. Acoust. Soc. Am. 131 (4), 342–347. Song, Aijun, Abdi, Ali, Badiey, Mohsen, Hursky, Paul, 2011. Experimental demonstration of underwater acoustic communication by vector sensors. IEEE J. Ocean. Eng. 36 (3), 454–461. Stojanovic, M., Jul. 2008. Efficient processing of acoustic signals for high rate information transmission over sparse underwater channels. Elsevier J. Phys. Commun. 146–161. Stojanovic, M., Catipovic, J.A., Proakis, J.G., Jan. 1994. Phase-coherent digital communications for underwater acoustic channels. IEEE J. Ocean. Eng. 19 (1), 100–111. Yang, T.C., 2007. A study of spatial processing gain in underwater acoustic communications. IEEE J. Ocean. Eng. 32 (3), 689–709. Yang, T.C., Kevin D, Heaney, Network-assisted Underwater Acoustic Communications, WUWNet'12, Los Angeles, CA, USA, Nov. 5-6, 2012. Zhang, G.S., Dong, H.F., 2011. Spatial diversity in multichannel processing for underwater acoustic communications. Ocean. Eng. 38 (14), 1611–1623.
Acknowledgments This work was supported in part by the National Key R&D Program of China under Grant 2016YFC1400200, the National Natural Science Foundation of China (Grant No. 61471298), and the Fundamental Research Funds for the Central Universities(Grant No.3102016AXXX03). The authors would like to thank all the participants of the lake experiment. The authors would also like to thank the anonymous reviewers for their valuable comments and suggestions on the original manuscript.
214