High Throughput Receiver Structure for Underwater Communication

Abstract

Acoustic channels are characterized by long multipath spreads that cause intersymbol interference. The way in which this fact influences the design of the receiver structure is considered in this paper. To satisfy performance and throughput requirements, we propose a consecutive iterative BCJR equalization scheme. To achieve a low error performance, we resort to the powerful BCJR equalization algorithms to iteratively update probabilistic information between inner decoder and outer decoder. Also, to achieve a high throughput, we divide a long packet into a group of small consecutive packets, estimate channel information of the current packets, and use it to decode the next packets. Based on an experimental channel response, we confirm that the performance is indeed improved for long packet size.

1. Introduction

It is well known that underwater channels are often hostile for underwater (UW) sensor communications, which impose three major obstacles for coherent transceivers. One is the excessive multipath delay spread in a UW channel, which usually causes the intersymbol interference (ISI). Another obstacle is the Doppler shift due to the relative motion between the source and the receiver, which causes compression or dilation on the received signals. The last one is the fast time varying phase drift due to random nature of the UW acoustic channels.

Various methods to cope with the multipath effect have been developed. A well-known method to counteract ISI is the decision feedback equalizer (DFE), which has been used in many UW sensor communication applications [1, 2]. However the use of DFE has difficulties when the multipath with a number of arrivals has equal strength or low SNR [3]. The other way to cope with ISI is to use an iterative equalizer which consists of an outer loop in addition to the inner loop BCJR decoder in the receiver. The assembly utilizes the error correcting capability of the convolutional codes to get an efficient equalizer [4, 5]. Alternatively, to cope with multipath effect, we adjust the packet length according to the channel coherence time. Due to the very short coherence time, we can only transmit small packets, which reduces the throughput. To achieve a high throughput, we divide a long packet into a group of small consecutive packets and use the estimated channel information of previous packets to compensate for the current and next packets. In this paper, we employ an iterative receiver structure with fine-tuned parameters to process experimental data from a fixed source to a fixed receiver at the data transmission rate of 1 k-symbol/s. The results indicate that the proposed algorithm works effectively well and indicate how much coding gains can be obtained as the iteration number increases.

2. Generic Transmission Model of Underwater Communication

The baseband model of the generic transmission system is shown in Figure 1. It shows that the iterative linear equalizer is a decision feedback equalizer, which deals with the outer code of the receiver. An inner code consists of convolutional codes. The information data to be transmitted are encoded by one-half rate convolutional codes with the constraint length of 7.

Figure 1

Baseband model of transmission system with turbo equalization.

Convolutional codes are used extensively in numerous applications in order to achieve reliable data transmission, including both wireless and underwater communication. A popular convolutional code with a constraint length of 7 and a rate of 1/2 is widely used. For example, the $(\mathrm{2,1},7)$ convolutional code with the generator matrix of ${(\mathrm{171},\mathrm{133})}_{\text{octal}}$ is used in [6]. The equalizer and the BCJR decoder are connected through the interleaving and deinterleaving functions that update each other's information recursively. The inner coded bits are then subtracted from the input and interleaved. The interleaved output is canceled a posteriori from the proceeding received signal. The interleaving function helps the receiver convergence. Generally, the decoding method of the convolutional code is a Viterbi decoding using a hard decision value. However, to improve the receiver performance through the iterations with DFE, a soft decision value is required. Therefore, we use the BCJR algorithm with the soft decision value. The BCJR algorithm is a well-known maximum a posteriori probability decoding algorithm, which has been proposed earlier for point-to-point communication applications [7].

Let $L_e^I$ be the output value of DFE as the estimated extrinsic value from the received signal. Let $y[k]$ be the equalizer input at time k; then the output of the DFE at time k, denoted by $L_e^I[k]$ , is given by

$$\begin{array}{cc}& L_e^I\left[k\right]={\displaystyle \sum }_{i=\mathrm{0}}^{N_b-\mathrm{1}}{c}_i\left[k\right]y\left[k-i\right]-{\displaystyle \sum }_{j=\mathrm{0}}^{N_a}{b}_j\left[k\right]{\widehat{L}}_e^I\left[k-j\right],\end{array}$$

(1)

where $c_i[k]\hspace{0.17em}\hspace{0.17em}(i=\mathrm{0,1},\dots ,N_{c-1})$ are the forward equalizer taps at time k, $b_j[k]\hspace{0.17em}\hspace{0.17em}(j=\mathrm{0,1},\dots ,N_b)$ are the feedback taps at time k, and ${\widehat{L}}_e^I[k]$ is the slicer output, which is the constellation point closest to $L_e^I[k]$ , as in [8, 9]. $e_s[k]$ is the Sato error given by

$$\begin{array}{cc}& e_s\left[k\right]=L_e^I\left[k\right]y\left[k\right]-\gamma \mathrm{sgn}\left\{L_e^I\left[k\right]\right\},\end{array}$$

(2)

where, γ is a constant value and $\mathrm{sgn}\left\{·\right\}$ is a sign function defined by

$$\begin{array}{cc}& \mathrm{sgn}\left\{x\right\}=\left\{\begin{array}{cc}-\mathrm{1}\hfill & \text{when\hspace{0.17em}\hspace{0.17em}}x<\mathrm{0}\hfill \\ 0\hfill & \text{when\hspace{0.17em}\hspace{0.17em}}x=\mathrm{0}\hfill \\ +\mathrm{1}\hfill & \text{when\hspace{0.17em}\hspace{0.17em}}x>\mathrm{0}.\hfill \end{array}\right.\end{array}$$

(3)

The value of $L_e^D$ after the interleaver is computed as $L_e^I-L_c^I$ and then input to the turbo decoder. The estimated extrinsic value of $L_c^D$ at the decoder output is given by

$$\begin{array}{cc}& L_c^D=\log \frac{P\left(x=+\mathrm{1}\right)}{P\left(x=-\mathrm{1}\right)}.\end{array}$$

(4)

The extrinsic value $L_c^D$ is an error correction term, from which the post probability is calculated. The reinterleaving of the computed value, $L_c^D-L_e^D$ , is input to the DFE; then $L_c^I$ is updated in order to compensate for the errors.

2.1. General Packet Structure

The packet structure is shown in Figure 2. A packet is preceded by a linear frequency modulation beginning (LFMB) signal to indicate that the packet is about to start. After silence period, the packet starts. The packet consists of preamble (P) and payload (D) fields. The preamble of 256 symbols is used for the initial channel and Doppler estimation. The payload field contains 500 coded QPSK symbols. The packet ends with another LFM signal identified as LFME (LFM End). During transmission, to avoid interpacket interference, the packet size should be small and the gap sufficiently long. These requirements apparently reduce the throughput.

Figure 2

General packet structure (change the data symbol length from 1000 to 500).

Figure 3 shows the procedure of using the preamble to assist the demodulation of the data symbols. Using the preamble data, the frequency and phase offset are estimated with a Doppler frequency and phase estimation algorithm, which will be described in Section 2.2. The estimated frequency and phase offset are fed to the payload field to compensate for the Doppler frequency and phase offsets. After the whole data packets are compensated using the estimated frequency and phase offsets, the compensated data are fed to the equalizer.

Figure 3

Frequency and phase offset compensation.

2.2. Doppler and Phase Estimation Algorithm

A preamble of known symbols is inserted to the beginning of each packet for the carrier frequency and clock timing recovery. Among various methods, data-aided (DA) estimations are normally employed to attain a good performance with a short preamble [10].

Filtering the received waveform in a matched filter and sampling it at proper times yield

$$\begin{array}{cc}& x\left(k\right)=d_k{e}^{j\left(\mathrm{2}\pi f_{d}kT+\theta \right)}+n\left(k\right),\end{array}$$

(5)

where $d_k$ is a unit-amplitude symbol, $f_d$ is the carrier frequency, and θ is the carrier phase. T is the symbol duration. $n(k)$ is a complex-valued independent Gaussian random variable. In the data-aided mode, the modulation can be easily stripped by multiplying $x\left(k\right)$ with $d_k^{\ast }$ (complex conjugate of $d_k$ ), which yields

$$\begin{array}{l}z\left(k\right)=d_k^{\ast }{d}_k{e}^{j\left(\mathrm{2}\pi f_{d}kT+\theta \right)}+n\left(k\right)d_k^{\ast }\\ \hspace{1em}=e^{j\left(\mathrm{2}\pi f_{d}kT+\theta \right)}\left(1+\tilde{n}\left(k\right)\right).\end{array}$$

(6)

As seen in (6), $z\left(k\right)$ can be viewed as a complex sinusoid embedded in the white Gaussian noise. A data-aided carrier frequency and phase estimator is derived based on the observation of a few consecutive samples $\left\{z\left(k\right),0\le k\le L-1\right\}$ . For the purpose of large frequency offset estimation, the following improved algorithm can be employed:

$$\begin{array}{cc}& {\widehat{f}}_d=\frac{\mathrm{1}}{\mathrm{2}\pi T}\arg \left\{{\displaystyle \sum }_{m=\mathrm{1}}^{N-\mathrm{1}}\left[R\left(m+\mathrm{1}\right){R\left(m\right)}^{\ast }\right]\right\},\end{array}$$

(7)

where $R(m)$ is the sample correlation function, given by

$$\begin{array}{l}R\left(m\right)=\frac{\mathrm{1}}{L-m}{\displaystyle \sum }_{k=m}^{L-\mathrm{1}}z\left(k\right)\times z^{\ast }\left(k-m\right)\\ m=\mathrm{1,2},\dots ,N.\end{array}$$

(8)

3. Proposed High Throughput Receiver Structure

To improve the system throughput, a new packet structure shown in Figure 4 is proposed. The lengths of P and $D_k$ are the same as those of the conventional packet structure shown in Figure 2. The only difference is that the length of the data field is doubled, which allows us to significantly improve the throughput for the same kind of overhead.

Figure 4

Packet structure.

This packet structure may suffer from performance degradation in the fast time varying channel. However, the performance degradation can be alleviated by a recursive frequency offset compensation procedure shown in Figure 5. In Figure 5, the 2-step demodulation procedure is illustrated: In the first step, the first data packet is demodulated using the Doppler and phase information estimated from the preamble, and in the second step, the second data packet is demodulated using the Doppler and phase information estimated from the first data packet after being demodulated. In fact, we can use this approach to divide a long packet into small consecutive packets and to demodulate the subsequent small packets using the information obtained from the previous small packets, thereby improving the system throughput in the fast time varying channel environments.

Figure 5

A recursive procedure of Doppler and phase compensation.

The frequency and phase offset information resulting from the preamble data is used to compensate for any frequency and phase offsets in the $D_{\mathrm{1}}$ packet. After the Doppler and phase offset are compensated, the $D_{\mathrm{1}}$ packet can be decoded through turbo equalization. Based on the demodulated $D_{\mathrm{1}}$ and reencoded and remapped $D_{\mathrm{1}}$ packet, we can estimate the frequency and phase offset that occurred in the $D_{\mathrm{1}}$ packet period and apply it to the $D_{\mathrm{2}}$ packet. The reencoded and remapped $D_{\mathrm{1}}$ packet acts as the “preamble” data of the $D_{\mathrm{2}}$ packet.

The receiver structure for the proposed high throughput packet model is shown in Figure 6. As explained above, to achieve high throughput, we divide a long packet into small consecutive packets and use the estimated channel information from the previous packets to assist the demodulation of the next packets. The proposed receiver consists of buffer, reencoder, and remapper. Buffer stores received packets in order to estimate the Doppler and phase offsets for the current data packet. For example, to estimate ${\widehat{f}}_d$ , θ from the $D_{\mathrm{1}}$ packet in order to demodulate the $D_{\mathrm{2}}$ packet, the received $D_{\mathrm{1}}$ packet has to be buffered. After estimating ${\widehat{f}}_d$ , θ, the previous packets are removed from the buffer. For the previous packets to be used as the preamble data for the current packet, reencoder and remapper are necessary. Let us assume that ${\widehat{D}}_k$ is output of the reencoder and remapper for the received $D_k$ packet; Doppler and phase estimation block determines ${\widehat{f}}_d$ , θ from $D_k$ and ${\widehat{D}}_k$ . The estimates ${\widehat{f}}_d$ and θ from $D_k$ are then used to demodulate $D_{k+\mathrm{1}}$ .

Figure 6

Proposed high throughput receiver model.

4. Experimental Results

We evaluate the performance of the proposed packet transmission model in real underwater environments. The experiment was conducted on a lake of Munkyeong city, Korea, in May 2014. The water depth was approximately 43 meters. Figure 7 depicts the lake trial.

Figure 7

Illustration of the lake trial.

The transmitter emits a source signal with center frequency of 16 kHz and data rate of 1 kbps. The hydrophone was at 20 m below the water surface, and the horizontal range from transmitter was 400 m. The received signal was sampled at 192 kHz. The experimentation with two types of packet structure is conducted.

In Figure 8, the underwater channel response is shown during a 0.1-second interval. This response was measured by using an LFM signal with a bandwidth of 4 kHz. The channel gains for the secondary and third arrivals fluctuate more rapidly. This sparse channel is affected by multipath propagation by reflection from the surface and the bottom. Experiment parameters are listed in Table 1.

Table 1

Parameters of UWA channel experiment.

Source	500-bit text
Channel coding	(2, 1, 7) convolutional code
Modulation	QPSK
Packet size ( $$D_k$$ )	500 symbols
Bit rate	1 kbps
Center frequency	16 kHz
Sampling frequency	192 kHz
Distance	400 m
Water depth	43 m
Depth	TX: 2 m, RX: 20 m

Figure 8

Underwater channel response.

Figure 9 illustrates the error performance of the conventional system and the proposed system. The received symbols have a total of 1047 bit errors in the first and second data fields ( $D_{\mathrm{1}}$ and $D_{\mathrm{2}}$ ) before any error correction. We investigate four different types of decoding methods: The first one is the hard decision Viterbi decoder, the second one is the soft decision Viterbi decoder, the third one is the conventional turbo equalization described in Section 2 (see Figure 3), and the last one is the proposed receiver structure of Figure 5. The first and second decoders have 619 bit errors and 492 bit errors, respectively. For the third decoding method, 451 bits errors are recorded after the first iteration, 408 bit errors are recorded after the second iteration, and finally the number of bit errors is reduced to 252 after the third iteration. In comparison, the proposed method can correct all the errors after three iterations. From the experimental results, we have confirmed that the proposed algorithm is very effective in improving the communication system performance for the fast time varying underwater communication channel.

Figure 9

BER comparison between the conventional and the proposed methods.

5. Conclusions

Underwater acoustic channels are characterized by long multipath spreads that cause intersymbol interference in addition to Doppler shift due to the relative source-receiver motion. To improve the system performance and throughput in the presence of these major obstacles, we have proposed a consecutive iterative BCJR equalization for a long packet size. To achieve the high throughput in the fast time varying channels, we divide the long packet into small consecutive packets and use previous packets to estimate channel information needed to compensate for the subsequent packets. The error performances of four different decoding algorithms are compared based on real experimental data measured on a lake in Munkyeong city. We demonstrate that the proposed algorithm can correct all errors of received symbols, thereby confirming that the proposed algorithm is very effective for communications in the time varying underwater communication channel.