Multi-Adaptive Filter Based On FPGA

Somatosensory evoked potentials (SEP) is an electrophysiological response to external stimulation, reflecting the integrity of the neurological pathway. SEP measurement is an essential test in spinal intraoperative monitoring.In clinical practices, SEP signal is usually accompanied by different kinds of noises, which make extremely poor signal-to-noise ration (SNR) during recording. Therefore, it is a key issue in signal processing to extract the SEP signal from the strong noise background.Conventional ensemble averaging (EA) is a commonly used technique in SEP processing to improve the SNR. However, it requires a huge number of inputs in order to extract a reliable SEP, and lack of the time varying information. The time delay of SEP detecting may result in permanent damage to the spinal cord. To solve this problem, this study develops a real time SEP extraction technique, which combines advanced signal processing algorithm with VLSI digital processing technique.This paper presents an algorithm combining multi-adaptive filter (MAF) with radial basis function neural network (RBFNN). The raw SEP is first denosied by adaptive noise canceller (ANC). Afterward, its output and the reference signal calculated by RBFNN are imported into the adaptive signal enhancer (ASE). The radial basis function here replaces the conventional EA signal so as to save time, retain variation and match the time-varying characteristics of SEP.The real time SEP extraction algorithm is implemented in the field programmable gate array (FPGA). For better performance of the hardware, the fixed point algorithm is developed. This paper analyzes the fixed point quantization effect of ANC, RBFNN and ASE, and optimizes the parameters of filters in fixed-point algorithm, such as stepsize, hidden nodes, and bit length. The optimal fixed-point algorithm is implemented in FPGA, and gets a good performance in extracting the real time SEP signal.