Low Complexity Wavelet Compression of Multichannel Neural Data
Krzysztof Duda, Paweł Turcza, Zbigniew Marszałek · 2018
In this paper, the integer-to-integer Lifting Wavelet Transform (LWT) is investigated in application to compression of 8 channels neural data sampled at 20 kHz with 12-bit resolution. The LWT operates on integers, and all multiplications can be implemented as bit shifts. The signal is transformed in blocks of 8 by 8 samples. Each block is decomposed on three levels of the LWT decomposition. Wavelet coefficients are next quantized, and finally, entropy encoded. The performance of the LWT is compared with the Lifting Wavelet Packet Transform (LWPT), and the Discrete Cosine Transform (DCT). For similar quality the DCT offers the highest compression, however, the computational complexity of the LWT is significantly lower, and also the LWT can operate in lossless mode.