Design and Implementation of an OFDM-based Voice Transmission System for
Mobile Underwater Vehicles
Seyyed Mahmood Jafari Sadeghi Mostafa Derakhtian Mohammad Ali Masnadi-Shirazi
School of Electrical and School of Electrical and School of Electrical and Computer Engineering Computer Engineering Computer Engineering Shiraz University Shiraz University Shiraz University Shiraz, Iran Shiraz, Iran Shiraz, Iran mjafari@shirazu.ac.ir derakhtian@shirazu.ac.ir masnadi@shirazu.ac.ir
Abstract—In this paper, we present an Orthogonal Fre-
quency Division Multiplexing (OFDM) based system, which is designed and implemented to transmit voice signal, over UnderWater Acoustic (UWA) channels. The system is software- based and using two Laptops, it would be possible to transmit and receive digital voice between two Underwater Vehicles (UV). System parameters are tuned for underwater acoustic channel observed in the Persian Gulf. The system may be used for speeds, up to 30 nautical miles per hour. Eventually some Figure 1. Block diagram of transmitter experimental results are presented that show the power of the system to achieve the desired performance. Keywords-OFDM system, Voice Transceiver, Underwater (employed in numerous wireless standards such as: IEEE Acoustic channel. 802.11a), researchers came with the idea of applying the OFDM technique in UWA channels [5]. I. I NTRODUCTION We have designed and implemented a system, based on Since electromagnetic waves do not propagate well in OFDM modulation technique for the UWA channel of the conductive environment of sea water, the best choice for Persian Gulf. In section II, the structure of the transmitter digital data transmission is to use acoustic waves. Use of will be explained, in section III, some details about receiver acoustic waves to transmit data in underwater channels, algorithms are presented, in section IV, the user interface has been under research for more than three decades [1]. of the system is introduced, in section V, the experimental This transmission method causes several issues that eventu- results are represented and finally concluding remarks are ally reduce the data rate of underwater acoustic channels, added in section VI. significantly. Most of the problems in underwater acoustic communication result from the very low traveling speed of II. T RANSMITTER S TRUCTURE the acoustic waves (1500m/s) in comparison with electro- Figure 1, shows the structure of the transmitter. Available magnetic waves. Large time spreads and wide-band Doppler bandwidth of the system is restricted to the bandwidth of the effects, Doppler scale, are two of these problems [3]. In acoustic transducers. The bandwidth is just 4-KHz centered addition, the wind causes variations in surface of the water, at the carrier frequency of 8-KHz. Sampling frequency is which leads the underwater acoustic channel to be a fast- chosen 48-KSPS which is the standard sampling rate of fading channel [3]. nearly all Laptop and PC sound-cards. In 1980s, researchers used orthogonal non-coherent mod- ulation schemes like the Frequency Shift Keying (FSK) A. Voice Compression and differentially coherent modulation schemes like the The Standard sampling rate for a voice signal is 8000 Differentially Phase Shift Keying (DPSK), in underwater samples per second, with resolution of 8-bits, which requires acoustic communications [1]-[2]. Use of coherent modula- data rate of 64-Kbps for transmission. Considering the tion schemes was difficult due to fast phase rotations of available bandwidth which is just 4-KHz, data transmission the UWA channel until 1994, when an adaptive equalizer with the rate of 64-Kbps requires very high power which integrated with Phase-Locked Loop (PLL) was proposed in is practically impossible. So we used the Mixed Excitation [4], to track phase rotations of the UWA channel. After the Linear Prediction (MELP) 2.4-Kbps voice coder to reduce OFDM technique was proposed in wireless communication the data rate of transmission from 64-Kbps to 2.4-Kbps [6].
Figure 3. Subcarrier assignment of each OFDM symbol
subcarriers) are inserted, which are exploited to estimate the
Compression Ratio (CR) of the MELP algorithm is nearly Carrier Frequency Offset (CFO) at the receiver side. 27. Data rate required to transmit output data of the vocoder is 2400 bps, which is acquirable using the 4-KHz bandwidth. E. PAPR Reduction B. Channel Coding One of the main disadvantages of OFDM technique is its To reduce overall BER of the system, the convolutional high Peak to Average Power Ratio also known as PAPR. coding of rate 12 with generator polynomial [171 133] In this system, we used the clipping method to reduce the and the coding depth of 7 is used. At the receiver side, amount of PAPR [10]. The optimum clipping threshold is Viterbi decoding algorithm with trace-back length of 34 derived via simulation. is used to recover data bits. Soft decoding of data bits III. R ECEIVER S TRUCTURE using Approximate Log-Likelihood Ratio (ALLR) algorithm is considered which has better performance than the hard Figure 4 describes the structure of the receiver. All of the decoding algorithm [7]. processing of the receiver (except the signal preamplifier and the transducer) are just software algorithms operating C. Data framing in real-time. Thus, the required hardware are just a Laptop In this system, the data is transmitted in the form of frames or PC, transducers and signal amplifiers. (data bursts), with duration of 765 milli-seconds each. Data transmission in the form of frames has two benefits: First, A. Timing synchronization and the Doppler scale estimation it facilitates the timing synchronization at the receiver side, Timing synchronization is the determination of the start second, the the Doppler scale can be estimated by knowing of frame and consequently the exact timing position of the the frame length sent by the transmitter and estimating the OFDM symbols. In this system, the timing synchronization frame length at the receiver [11]. Frame structure of the is achieved by correlating the input samples with local copy system is depicted in Figure 2. of the HFM chirp signal. The metric M (d) defined in (1), is As shown in Figure 2, a Hyperbolic Frequency Modulated calculated over the input samples. Whenever M (d) exceeds (HFM) Chirp signal of length 17ms is inserted at the a predefined threshold, the existence of a frame is declared beginning of each frame [9]. The chirp signal marks the and the peak of M (d), shows the exact beginning of the beginning of the frame for the receiver. After the chirp frame. ∑N −1 signal, a null time is inserted to avoid interference of the | m=0 y[d + m]S ∗ [m]| chirp signal with the other portions of the frame. This time M (d) = √∑ (1) N −1 2 should be greater than the time-spread of the channel. Data m=0 |y[d + m]| portion of each frame consists of five OFDM symbols. Each In (1), y[.] represents the received samples, S[.] is the OFDM symbol has a length of 128ms and 18ms zero-pad. known HFM Chirp signal and ∗ is the complex conjugate Zero-Pad (ZP) OFDM is used instead of conventional Cyclic operation. Because of the Doppler scale phenomenon in Prefix (CP) OFDM to avoid the dissipation of power in the mobile acoustic systems [3], the frame length at receiver CP portion [8]. is not equal to the frame length at the transmitter side. If D. Subcarrier assignment of OFDM symbols the Doppler scale is not compensated, performance of the Figure 3 demonstrates the assignment of subcarriers of system is reduced significantly due the wide-band nature each OFDM symbol. Each OFDM symbol consists of 512 of the Doppler effect. Here, we estimate and compensate subcarriers. 128 of these are used as pilots to estimate the the Doppler scale using the method described in [11]. An channel at the receiver. Pilot sequences are inserted equally- estimate of the Doppler scale can be calculated by: spaced to facilitate use of the Least Squares (LS) channel ˆ = Trx − 1 ∆ (2) estimation algorithm [14]. Also, 9 null subcarriers (virtual Ttx
ˆ is the estimate of the Doppler scale, Trx is the In (2), ∆ calculated frame length at the receiver and Ttx is the frame length at the transmitter which is 765-ms. The length of the received frame is extracted using the chirps of two consecutive frames. The Doppler scale is com- pensated using computationally efficient linear interpolation method [11]. B. CFO estimation As described in [12], inaccurate estimation of the Doppler scale parameter, causes the signal to suffer from narrow- band Doppler shift or in other terms CFO. If the CFO is not compensated, subcarriers would not be orthogonal anymore and this would cause an increase in BER. In our system, estimation of CFO is accomplished by exploiting the null subcarriers. The idea is to minimize the cost function derived in [13]: ∑ H 2 J(ϵ) = |ν H i P y| (3) i∈Γz Figure 5. Deployed Graphical User Interface of project ϵ̂ = arg min J(ϵ) (4) ϵ
where in (3), νi is a 512 × 1 vector with the nth element
j2π(n−1)i given by √5121 e 512 , P is 512 × 512 diagonal matrix, P = diag(1, ej2πϵ , ..., ej2π×511×ϵ ), y is the 512×1 vector of the samples of received OFDM symbol, H is the Hermitian operator and Γz is the subset of null subcarriers. In noiseless conditions, the cost function of the equation (3) which shows the energy of null subcarriers is minimized for the correct value of ϵ, so ϵ̂ defined in (4) would provide an estimate of ϵ. C. OverLap and Add Receiver Since, instead of conventional method of CP-OFDM, ZP- Figure 6. Location of transmitter and receiver on Google Earth program OFDM is used, another processing is required before the FFT operation. The method is called OverLap and Add and is described in [8]. The performance is reduced a little bit, to receive/transmit the samples of voice/modulated signal. since the optimal detection of ZP-OFDM is not used, but The GUI has the ability to select between different sound- the energy saving due to elimination of the redundant CP cards that are installed on the PC or Laptop. Also, the frame portion is significant. detection threshold, transmitter power and volume of the D. Channel Estimation voice signal are tunable via the GUI. We used the LS method described in [14] to estimate V. E XPERIMENTAL R ESULTS and compensate the channel. The LS channel estimation is simplified when the pilot symbols are equally-spaced and Some experiments were performed in the the Persian Gulf, can be accomplished efficiently by the use of Fast Fourier near Bandar Bushehr. Table I summarizes the parameters Transform (FFT) and Inverse FFT (IFFT) operations [14]. of the system used in the experiments. The exact positions of the transmitter and the receiver are depicted in the IV. G RAPHICAL U SER I NTERFACE (GUI) Figure 6. The position of the transmitter was : Latitude : The processing algorithms of the transmitter and the 2859’56.65”N Longitude : 5041’46.52”E and the position receiver are written in C language. But a graphical user of the receiver was Latitude : 29 1’15.36”N Longitude : interface has been designed by using the Visual C# lan- 5038’28.27”E .The transmitter and the receiver were ap- guage, to ease the use of the program. Figure 5 shows proximately 6-Km apart. The depth of water in that region the Graphical User Interface of the system. As mentioned is nearly 16 meters. The transducers’ depth were 3 meters. before, the program uses sound-card of the PC or Laptop Speed of the wind was 13 Km/h, at the time of experiment.
Table I S YSTEM PARAMETERS [2] G. R. Mackelburg, S. J. Watson, and A. Gordon, Benthic 4800 bps acoustic telemetry, presented at the Oceans81, Boston, Parameter Value MA, 1981. Bandwidth 4-KHz [3] Stojanovic, M.; Preisig, J.; , ”Underwater acoustic commu- Carrier frequency 8-KHz nication channels: Propagation models and statistical charac- Sample rate 48-KHz terization,” Communications Magazine, IEEE , vol.47, no.1, Total number of subcarriers 512 pp.84-89, January 2009. Number of data subcarriers 375 [4] Stojanovic, M.; Catipovic, J.A.; Proakis, J.G.; , ”Phase- Number of Pilot subcarriers 128 coherent digital communications for underwater acoustic chan- Number of null subcarriers 9 nels,” Oceanic Engineering, IEEE Journal of , vol.19, no.1, Number of OFDM symbols per frame 5 pp.100-111, Jan 1994. Subcarrier spacing ≈ 7.8 Hz [5] Hai Yan; Shengli Zhou; Zhijie Shi; Jun-Hong Cui; Lei Wan; Jie Frame length 765ms Huang; Hao Zhou; , ”DSP implementation of SISO and MIMO Frame marker length (chirp signal) 17ms OFDM acoustic modems,” OCEANS 2010 IEEE - Sydney , Duration of OFDM symbol 128ms vol., no., pp.1-6, 24-27 May 2010. Gaurd intervals 18ms [6] McCree, A.V.; Barnwell, T.P., III; , ”A mixed excitation LPC vocoder model for low bit rate speech coding ,” Speech and Table II Audio Processing, IEEE Transactions on , vol.3, no.4, pp.242- T EST C ONDITIONS 250, Jul 1995. Condition Value [7] Viterbi, Andrew J., ”An Intuitive Justification and a Simpli- Horizontal spacing ≈ 6-Km fied Implementation of the MAP Decoder for Convolutional Distance from the coast line ≈ 10-Km Codes,” IEEE Journal on Selected Areas in Communications, Depth of water 16-m Vol. 16, February 1998. Depth of transducers 3-m [8] Muquet, B.; Zhengdao Wang; Giannakis, G.B.; de Courville, Wind speed 13-Km/h M.; Duhamel, P.; , ”Cyclic prefixing or zero padding for wireless multicarrier transmissions?,” Communications, IEEE Transactions on , vol.50, no.12, pp. 2136- 2148, Dec 2002. The approximate distance of the transmitter and receiver [9] Feng Wei; Xu Xiaomei; Zhang Lan; Chen Yougan; , ”A with the coast line was 10-Km. We observed the coded frame synchronization method for underwater acoustic com- BER of < 10−4 for calm weather conditions, but when the munication on mobile platform,” Image Analysis and Signal wind speed was increased to 20 Km/h, the BER became Processing (IASP), 2010 International Conference on , vol., approximately 10−2 , which slightly reduced the quality of no., pp.518-522, 9-11 April 2010. the received voice (The increase in BER is due to Doppler [10] Xiaodong Li; Cimini, L.J., Jr.; , ”Effects of clipping and spread effect). Table II summarizes the test conditions of the filtering on the performance of OFDM,” Vehicular Technology system. Conference, 1997, IEEE 47th , vol.3, no., pp.1634-1638 vol.3, 4-7 May 1997. VI. C ONCLUSION [11] Sharif, B.S.; Neasham, J.; Hinton, O.R.; Adams, A.E.; , ”A computationally efficient Doppler compensation system for We presented a fully digital system for transmission of underwater acoustic communications,” Oceanic Engineering, voice over the underwater acoustic channels. The system IEEE Journal of , vol.25, no.1, pp.52-61, Jan 2000. is software-based which means the system is a computer program and can be installed on any PC or Laptop. Using [12] Baosheng Li; Shengli Zhou; Stojanovic, M.; Freitag, L.; this program, voice communication between two underwater Willett, P.; , ”Multicarrier Communication Over Underwater Acoustic Channels With Nonuniform Doppler Shifts,” Oceanic vehicles is possible at a low cost and acceptable perfor- Engineering, IEEE Journal of , vol.33, no.2, pp.198-209, April mance. The experimental Results in the Persian Gulf showed 2008. proper performance of the system in moderate weather conditions. [13] Defeng Huang; Letaief, K.B.; , ”Carrier frequency offset estimation for OFDM systems using subcarriers,” Communica- tions, IEEE Transactions on , vol.54, no.5, pp. 813- 823, May R EFERENCES 2006. [1] D. J. Garrood, ”Applications of the MFSK acoustical com- [14] Li, B.; Zhou, S.; Stojanovic, M.; Freitag, L.; , ”Pilot-tone munication system,” presented at the Oceans’81, Boston, MA, based ZP-OFDM Demodulation for an Underwater Acoustic 1981. derwater communications,” presented at the Oceans’81, Channel,” OCEANS 2006 , vol., no., pp.1-5, 18-21 Sept. 2006. Boston, MA, 1981.