International Journal of Engineering Trends and Technology (IJETT) – Volume 19 Number 4 – Jan 2015 FPGA Implementation of Adaptive Filtering Algorithm for Noise Cancellation in Speech Signal Daphni S#1, Bamini S#2, Judith Johnsi J#3, Thavasumony D#4 #1 Assistant Professor, Department of ECE in Satyam College of Engineering, Midalakkadu, India. #2 Student, ME –Applied Electronics in Satyam College of Engineering, Velliavilai, India. #3 Part time Lecturer, Working in Muscat College, Sultanate of Oman, Changai, India. #4 Lecturer, Working in Ambo University Ethiopia, Midalakkadu, India. Abstract— Noise reduction of speech signals is a key challenge problem in speech enhancement, speech recognition and speech communication applications, etc. It has attracted a considerable amount of research attention over past several decades. The most widely used method is optimal linear filtering method, which achieves clean speech estimate by passing the noise observation through an optimal filter or transformation. Most common problem in speech processing is the effect of interference noise in speech signals, Interference noise masks of the speech signal and reduces its Intelligibility. It is necessary to remove the noise from the speech signals to get the clear understanding of the information that the speech signal contains. Normally, LMS adaptive filter is used for the process of noise removal in the speech signals. The Direct Form LMS adaptive filter is the most popular and most widely used adaptive filter, not only because of its simplicity but also because of its satisfactory convergence performance. In order to achieve the above mentioned objective, the concept of adaptive filtering algorithm is to be used. This algorithm is developed using MATLAB version 7.8.0.347(R2009a) and Xilinx 9.1 MSE, the comparison is done with LMS and NLMS Algorithms. Keywords— FPGA, Adaptive Filtering Algorithm, DSP, Least Mean-Square (LMS), MSE, Noise Cancellation. I. INTRODUCTION A filter is a device or process that removes some unwanted component from a signal. The drawback of filtering is the loss of information associated with it. Signal combination in Fourier space is an alternative approach for removal of certain frequencies from the recorded signal. Adaptive filtering techniques must be implemented to promote accurate solutions and a timely convergence to that solution. Adaptive digital filters have a wide range of communication and DSP applications such as adaptive equalization, system identification and image restoration and noise removal. The most widely used algorithm for adaptive filters is the least mean-square (LMS) algorithm due to its superior performance and simple calculation. The LMS algorithm is well suited to software-based simulation and analysis but is not applicable to hardware implementation. The LMS adaptive algorithm minimizes approximately the mean-square error by recursively altering the weight vector at each sampling instance. A. Speech Signal Speech is an acoustic waveform that conveys information from a speaker to a listener. Given the importance of this form of communication, it is no surprise that many applications of ISSN: 2231-5381 signal processing have been developed to manipulate speech signals. Speech sounds can be divided into two broad classes according to the mode of excitation. The two classes are voiced sounds, unvoiced sounds. At a linguistic level, speech can be viewed as a sequence of basic sound units called phonemes. The same phoneme may give rise to many different sounds or allophones at the acoustic level, depending on the phonemes which surround it. Different speakers producing the same string of phonemes convey the same information yet sound different as a result of differences in dialect and vocal tract length and shape. The speech signal has certain properties: It is a onedimensional signal, with time as its independent variable, it is random in nature, it is non-stationary, and i.e. the frequency spectrum is not constant in time. Although human beings have an audible frequency range of 20Hz to 20 kHz, the human speech has significant frequency components only up to 4 kHz. The most common problem in speech processing is the effect of interference noise in speech signals. Interference noise masks the speech signal and reduces its intelligibility. But low-frequency noise is a much more effective mask when the noise is louder than the speech signal, and at high sound pressure levels it masks both vowels and consonants. B. Objective of this Project The main objective of this thesis is to investigate the implementation of a real time noise cancellation application. The real time implementation has been carried out by a spartran-3E kit. First, the LMS, NLMS and AFA algorithms are simulated using MATLAB. Then, these algorithms are transferred to the spartran-3E kit which let them to work alone in real time independent of MATLAB. Furthermore, the performances of the algorithm can be measured by setting the appropriate speech signal data. II. LITERATURE REVIEW A. An Embedded Adaptive Filtering System on FPGA Changchoo, Prasannalakshmi Padmanabhan, SusmitaMutsuddy, proposed NLMS adaptive filter on FPGA; filter is an essential part of many DSP systems including echo cancellers, channel equalizers and noise cancellers. In this paper, we describe an embedded NLMS adaptive filtering system consisting of a flexible LMS core and Xilinx MicroBlaze soft processor, both of which are implemented on a Xilinx Spartan-3 FPGA. The LMS adaptive filter core, which features parameterizable bit-width and tap-length, has http://www.ijettjournal.org Page 190 International Journal of Engineering Trends and Technology (IJETT) – Volume 19 Number 4 – Jan 2015 been coded in Verilog-HDL, simulated with ModelSim and synthesized and implemented using Xilinx ISE tools. The normalization algorithm is coded in C and runs on the MicroBlaze processor. This embedded system has been implemented in a Xilinx ‗Spartan-3 Starter Kit‘ board using Xilinx EDK toolset. This system works at a clock frequency of 50MHz and uses about 90% of XC3S200 Spartan-3 FPGA Slice resources. Moreover, FPGAs have become more flexible and scalable as they come with a processor embedded in them. For example, Xilinx Virtex-II Pro device has a Power PC processor embedded in it. There is also an option of embedding a soft processor core, such as MicroBlaze, in various Xilinx FPGAs that do not have the embedded hardcore processor. B. Performance Limitations of Acoustic Echo Cancellers for Hands free Telephony Long,G, Ling.F and Proakis,J.G,[14] proposed a simple transversal Finite Impulse Response (FIR) adaptive filter using the Normalized Least Mean Squares (NLMS) algorithm could achieve perfect cancellation. However, achieving a perfect model in a realistic environment is a difficult problem. The limitations addressed in this include: under modelling of the acoustic impulse response (AIR) of the room, room noise (fans, air conditioning), near end speech disturbance (double talk), and the ability of a particular algorithm to quickly converge and dynamically track a changing AIR while objects move inside the room. Hands free telephones (HFTs) however include many other components which are not usually accounted for in conventional AEC designs but need to be considered in order to achieve optimal results during the identification process. These include electronic circuit noise, finite precision and truncation effects that occur when the analog signal is processed in digital form, vibration and resonances in the plastic enclosure as the loudspeaker emits a signal, microphone mechanical vibration sensitivities (as opposed to acoustical sensitivity) and nonlinearities which can occur in the loudspeaker and signal amplifiers.A typical HFT is illustrated in and normally consists of two Adaptive Filters (AF). The first AF is used to remove acoustic echoes and the second AF is used for cancelling echoes from an imperfect hybrid as well as reflections from the line. Conventional AECs utilize a linear adaptive transversal filter to model the room impulse response and cancel the echo signal. The NLMS algorithm is the baseline by which performance of alternative models is measured but the linear transversal filter architecture is incapable of reducing (for example) nonlinear distortion and will almost always be in the under modelled state for a typical room acoustic impulse response. As a result, a revised echo path model is required which includes all of the above limitations. In this the relative seriousness of these limitations to the achievable echoes cancellation. We outline the relevant performance requirements according to the currently available standards, review the characteristics of reverberant rooms and including reverberation time and how this affects performance, and present some measurement procedures. ISSN: 2231-5381 III. ADAPTIVE FILTER AND LMS ALGORITHM A. Noise and its Classifications A disturbance that affects a signal and that may distort the information carried by the signal. One more definition says that, random variations of one or more characteristics of any entity such as voltage, current, or data. White Noise Colored Noise Impulsive Noise B. Effects of Noise Whether or not a sound is undesired by a person will depend on a number of factors: Loudness, Frequency, Continuity, Variation with time, Time of occurrence, Information content, Origin of the sound, Recipient's state of mind and temperament, Background noise level. In general, the effects of noises are: Hearing Loss Other Health Effects Speech Interference and Masking C. Adaptive Noise Cancellation The purpose of adaptive noise cancellation is to improve the signal-to-noise ratio (SNR) of a signal by removing noise from the signal that we receive. The adaptive cancellation configuration is shown in Fig.1. Fig.1 Adaptive Noise Cancellation Here x (n) – input N1 (n) – noise source d (n) – desired signal Also the signal s (n) is corrupted by N0 (n). D. Adaptive Filters There are four major types of adaptive filtering configurations; adaptive system identification, adaptive noise cancellation, adaptive linear prediction, and adaptive inverse system. All of the above systems are similar in the implementation of the algorithm, but different in system configuration. FIR filters are particularly useful for applications where exact linear phase response is required. The FIR filter is generally implemented in a non-recursive way which guarantees a stable filter. FIR filter design essentially consists of two parts approximation problem and realization problem. http://www.ijettjournal.org Page 191 International Journal of Engineering Trends and Technology (IJETT) – Volume 19 Number 4 – Jan 2015 The filter length of Adaptive system is inherently tied to many of the other performance measures. introducing the variance factor in the AFA Algorithm, which produces next sets of filter coefficients. E. Performance Measures in Adaptive Systems Convergence Rate Mean Square error Computational Complexity Stability F. Implementation of Adaptive Filtering Algorithm for speech signal on FPGA In this project implementation, the adaptive filtering algorithm (AFA) are implemented in MATLAB and also implemented in Verilog hardware description language. To evaluate the performance of the adaptive filtering algorithm and extensive simulations have been performed. The implementation process comprises of the following steps 1) To implement and evaluate the performance of adaptive filtering algorithm, noisy speech signal is passed through the adaptive Filter to remove the noise from signal. 2) Least Mean Squares (LMS), one of the widely used algorithms in many signals processing environment, is implemented for adaption of the filter coefficients. The cancellation system is implemented in MATLAB. The simulation design of adaptive filter is performed and analysed on the basis of Signal to Noise ratio (SNR) and Mean Square Error (MSE). 3) The adaptive filtering algorithm is implemented in time domain rather than in frequency domain. This is done to accommodate the random nature of speech signal. 4) The cancellation system is implemented in MATLAB and Verilog hardware description language. The simulation of MATLAB design of adaptive filter is performed and analysed on the basis of Signal to Noise ratio (SNR). IV. RESULTS AND DISCUSSION Fig.2 RTL Schematic Diagram of AFA Algorithm B. Simulation Results In this thesis, the modelsim Software Package is used for the simulation of the AFA algorithms in noise cancellation configurations. Simulations discuss the performances of these algorithms with white Gaussian noise with different parameters and different input signals. In the first part of this experiment, the step size is assumed to be μ = 0.006 for LMS, NLMS and AFA algorithms. The simulation results of LMS, NLMS, and AFA algorithm are shown in the following figures. The recorded sentence ―hello‖ was used as the clean speech. This sentence is conventionally used for speech processing. Hence the variability of effect of noise on speech with frequency of the signal is accounted. Simulation Result of LMS Algorithm: A. AFA Algorithm In the existing algorithm, improve the performance in terms of signal to noise ratio. By introducing a variance factor in the AFA algorithm, the performance of the algorithm cans be improved. The AFA algorithm uses an FIR filter structure. The main components of the filter consist of Filter block and L Weight Updates. The filter blocks are simply Flip-flops. Each weight Update component consists of a multiplier, an adder and a buffer to store the new weights‘ update of the filter coefficient. According to equation (2), the filter output is subtracted from the desired signal to produce an error signal. The RTL Schematic Diagram is given in Fig.2. The error signal is then multiplied with μ and then with the input signal, and by ISSN: 2231-5381 Fig.3 Speech and noise Signal In this section the speech signal added with noise for the fixed step size and original speech signal ―hello‖. White Gaussian noise was generated and added to the original speech signal. The original speech and noise signal is given in Fig.3. The SNR of the signal corrupted with noise. A linear http://www.ijettjournal.org Page 192 International Journal of Engineering Trends and Technology (IJETT) – Volume 19 Number 4 – Jan 2015 combination of the generated noise and the original signal is used as the primary input for the filter. Denoised signal can be recovered by th LMS algorithm with the appropriate weight update equation and also the fixed step size 0.006 and it s recover the same speech signal hello and white gaussian noise. The LMS, NLMS and AFA MSE Signals are shown in Fthe following figures. Also the AFA denoised signal is given in Fig.6. Fig.7 AFA MSE Signal Comparision result: The comparision has been done on the basis of signal to noise ratio and mean square error for the same speech signal ―hello‖ and the step size 0.006 has been fixed for all the algorithm and performance analysis can be shown in table 1 is illustrated. TABLE 1 PERFORMANCE ANALYSIS ON THE BASIS OF SNR AND MSE Fig.4 LMS MSE Signal Simulation Result of NLMS Algorithm: ALGORITHM SNR RATIO(in dB) MSE LMS 5.2416 2.8703e-07 NLMS 5.8490 1.1919e-07 AFA 6.7240 3.4661e-08 Verilog HDL simulation Result: Xilinx ISE 9.l development environment was used for implementation of above VLSI design. The design has been transferred to Verilog HDL code and its the hardware simulation done with the Xilinx ISE simulator and also implemented on Spartan-3 FPGA. Fig.5 NLMS MSE Signal Simulation Result of AFA Algorithm: Fig.8 Verilog Simulation of Adaptive Filtering Algorithm V. CONCLUSIONS Fig.6 AFA denoised Signal ISSN: 2231-5381 The most common problem in speech processing is the effect of interference noise in speech signals. Interference http://www.ijettjournal.org Page 193 International Journal of Engineering Trends and Technology (IJETT) – Volume 19 Number 4 – Jan 2015 noise masks the speech signal and reduces its Intelligibility. It is necessary to remove the noise from the speech signals to get the clear understanding of the information that the speech signal contains. Normally, LMS adaptive filter is used for the process of noise removal in the speech signals. The Direct Form LMS adaptive filter is the most popular and most widely used adaptive filter, not only because of its simplicity but also because of its satisfactory convergence performance. In recent signal processing domain we need to remove or cancel the noise which is added during signal passes through transmission medium. So we proposed novel hardware architecture for Adaptive Noise cancellation LMS filter. This filter follows sequence of steps to cancel unwanted signal. We developed our hardware design for adaptive digital filtering and it is developed by Verilog HDL. This design has been synthesized using Xilinx 9.1 ISE and we obtain simulation and synthesis results, using simulation we verified our expected results and then the result of synthesis has been targeted for Spartan-3E FPGA device and we got set of parameter. This parameter has been compared with previous work, after comparison we can conclude that we obtain optimal result. REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] AllamMousa, MarwaQados, Sherin Bader., (2012).― Speech Signal Enhancement Using Adaptive Noise Cancellation Techniques,‖ Canadian Journal on Electrical and Electronics Engineering Vol. 3, No. 7. Asit Kumar Subudhi, BiswajitMishra, Mihir Narayan Mohanty., (2011). ―VLSI Design and Implementation for Adaptive Filter using LMS Algorithm,‖ International Journal of Computer& Communication Technology (IJCCT), Volume-2, Issue-VI. Dimitris G. Manolakis,Vinay K. Ingle, Stephen M. Kogon.,(2005).― Statistical and Adaptive Signal Processing‖, McGrawHill, ISBN: 1-58053-610-7. Eriksson L.I, Allie M. C., and. Bremigan C. D., ( 2004). "Active Noise Control using Adaptive digital Signal Processing‖ in Proc. ICASSP , New York, pp. 2594-2597. L.R.Rabiner, R.W. Schafer.,(2004).―Digital Processing of speech signals‖, Pearson Education, ISBN 81-297-0272-X. Long,G,Ling,F and Proakis,J.G.,(1989). ―The LMS algorithm with delayed coefficient adaptation,‖ IEEE Trans. Acoust., Speech, Signal Process., Vol. 37, No. 9, pp. 1397–1405. Meher, P. K. andMaheshwari,M.,(2011), ―Low adaptation-delay LMSadaptive filter part-II: An optimized architecture,‖ in Proc. IEEE Int. Midwest Symp.Circuits Syst., pp. 1–4. Rocher,R, Menard,D ,Sentieys,O. and Scalart,P., (2004). ―Accuracy evaluation of fixed-point LMS algorithm,‖ in Proc. IEEE Int. Conf. Acoust., Speech,Signal Process., pp. 237–240. Sayed. A. Hadei, M. lot fizad J.,(2010).―A Family of Adaptive Filter Algorithms in Noise Cancellat ion for Speech Enhancement‖ International Journal of Computer and Electrical Engineering,Vol. 2 No.2. Shigeji Ikeda and Akihiko Sugiyama., (1999). ―An Adaptive Noise Canceller with Low Signal Distortion for Speech Codecs‖.ieee transactions on signal processing, Vol. 47, no. 3, March 1999. Simon Haykin.,(2008) "Adaptive Filters Theory" Pearson Education. Van, L, D. And. Feng, W, S., (2001). ―An efficient systolic architecture for the DLMS adaptive filter and its applications,‖ IEEE Trans. CircuitsSyst. II, Analog Digital Signal Process, Vol. 48, No. 4, pp. 359– 366. Widrow,B. and Stearns,S.D.,(2005). Adaptive Signal Processing. Englewood Cliffs, NJ, USA: Prentice-Hall. ISSN: 2231-5381 http://www.ijettjournal.org Page 194