Research On Optimization Of G.726Speech Coding And IP-core Design

PCM (pulse-code modulation) is the earliest and most widely used speech coding technology. At present, the adoption of non-linear quantization coding in64Kbit/s A-law and u-law PCM is widely applied to many communication systems. In1988, the speech coding standard of G.726ADPCM (adaptive differential PCM) was proposed and the coding rate standard of16Kbit/s,24Kbit/s,32Kbit/s,40Kbit/s channel was established by ITU(International Telecommunication Union). Recommendation G.726is characterized by comparative simple algorithm, higher-quality of speech restoration, guarantee of high speech quality after many times converted, and widely application to digital circuit multiplication equipment (DCME).In this thesis, Firstly, through analysising the models of ITU-T G.726standard in detail, the codecs based on ITU-T G.726standard was designed by the Visualc program. The codecs was divided into twe parts, one is encoder and the other is decoder. Decoder was designed according to G.726decoder, whereas the encoder module consists of two parts, an adaptive quantization module and an adaptive prediction module. In adaptive quantization algorithm, six sub-modules hed been implemented. They are Input PCM format conversion, difference signal computation, quantizer scale factor adaptation, adaptation speed control, tone and transition detector and adaptivequantizer. In adaptive prediction algorithm, three sub-modules are implemented.They are inverse adaptive quantizerand, adaptive predictor and reconstructed signal calculator. Then the speed test at16Kbit/s,32Kbit/s about this program had been proceed.Through this test, the one at16Kbit/s has better to restore the source audio.But it also has many sever problems such as Voice distortion, noise reduction.In order to solve the above problems, This thesis conducted a comparative analysis of waveforms firstly and found that compared with the source audio, the output bit-rate voice envelope has many mutations. This is because after using the adaptive prediction technique to remove the speech signal’s correlation, the reconstructed source signal at one end of the encoder can’t restore the relevance of speech signal accurately.That is the weakened linear correlation of the speech signal cause the voice distortion and noise reduction. Therefore, this thesis reconstructs the output speech from the decoder-side. By the use of backward recursion arithmetic average method, it enhanced speech signals’linear correlation, and carried out test experiments. Test results show that this method can solve the above problem at16kb/s bit-rate and through the experiments it proved that the average results when N=3is optimal, the signal noise ratio has increased3.54dB.With the development of EDA (Electronic Design Automation); technology, The advantage of using programmable logic devices FPGA((Field Programmable Gata Array)) to achieve real-time speech signal processing is more and more obvious, naturally the high-speed performance and reconfigurability of FPGA make up for the disadvantage of DSP (Digital Signal Processor) and dedicated ASIC chip.Besides the IP multiplexing technology will significantly reduce the development costs and development cycle of FPGA, and turn into the core areas of FPGA application and development. In this paper, based on the comprehension of Recommendation G.726, we develop the daptive prediction module in core part of this standard by soft IP technology. The IP core ased on the QUATUS II platform to develop with the VHDL (Hardware Description Language) language, and use modelsim simulation software to simulate and optimize, so that the target such as running speed, portability, resources using will achieve the best effect.