Published on Feb 21, 2020
Hi-Fi paper presents a sufficiently low bit rate Hi-Fi audio coding technique with low computation designed for transmitting real-time high-quality audio signal over wireless channel.
This technique applies wavelet packet transform to decompose audio signal into subbands to eliminate redundant data using spectral and temporal masking properties. The encoded audio data is framed with some critical field is protected by channel coding to improve noise immunity when frames are transmitted wirelessly. Experimental results show that transparent CD-audio quality can be achieved at 80kbps encoding bit rate. Moreover, the proposed technique still offers near CD-audio quality when frames are transmitted over AWGN channel with BER below 10-5. These encouraging results clearly exhibit the superior features of our technique compared to others such as Ogg/Vorbis and MP3, which are ubiquitously employed nowadays.
Current hi-fi audio CODECs employ entropy coding such as run-length and Huffman code, where important parameters required to decode are assumed to be error−free. Otherwise, the frame will be discarded making it susceptible to noisy wireless channel. As a result, we identify such parameters and protect them using channel coding that can correct up to 29 bits (one of every seven bits). Perceptual coding technique is used reduce the bit rate based on “human hearing masking” property. In general, perceptual codec consists of five modules which are filer bank, psychoacoustic analyzer, bit allocation, quantizing and encoding, and framing.
The output from filter bank (time/frequency analysis) is quantized according to masking thresholds calculated by the psychoacoustic analyzer. In [1,2] use polyphase filter bank which requires 512 coefficients to represent each filter. This can take significant time to encode the signal. In contrast, we use wavelet−based filter bank to transform signal into wavelet domain that analyzes both time and frequency simultaneously. The filter bank uses fewer coefficients and can represent variable sized subbands that more accurately match the characteristic of non−stationary audio signal ; human can only detect frequency difference at low− or medium−frequency.
This hypothesis leads us to believe that the proposed technique should be simpler (can be implemented in hardware). Moreover, MATLAB experiments confirm that, at 80kbps, our wavelet−packet audio codec yields comparable audio quality to that of the 64kbps MP3 and Ogg/Vorbis. It is worth noting that the higher bit rate of wavelet codec partially accounts for channel coding. This makes it wireless transmission ready.
Wavelet-packet Audio CODEC
This section explains structure and function of the proposed wavelet−packet codec which takes audio CD samples (44.1ksps @ 16bps PCM) as its inputs. Audio samples are framed at 1,024 samples each . Fewer samples would affect the coding efficiency whereas more samples would impose long coding delay, violating the real−time requirement. Subsequent frames share 16 overlapping samples to reduce discontinuity between reconstructed frames at the decoder. Each frame is windowed by raised cosine filter to avoid sudden change of the signal  then proceeds to the encoder