Compressive Sensing system for recording of ECoG signals in-vivo

5. System Level DesignCompressive Sensing is a novel compression method, and as it is discussed in the firstchapters, there are just few implementations oriented to wireless on-chip neural acquisition.Regarding to this, the assumptions have to be taken into consideration to develop a completemultichannel system are still in an early phase. As it has been introduced in Chapter 3, one ofthe main purposes of this thesis is to deepen into on the strengths and weaknesses of ananalog implementation in order to clarify if area, power consumption and reliability arecompetitive with the digital implementations, which are a more common approach in biosignalbasedapplications. From this starting point, the main parameters which have been assumed forthe Matlab and Cadence design are summarized in Table 5.1:ParameterSpecificationSamples (N) 128Measurement (M) 64Compression Ratio (C R ) 2Neural signalECoG and/or APBandwidth (BW)10 – 12 kHzMixing Frequency (f s )30 kHzTable 5.1. Main parameters of the CS analog design.The samples and measurement have been chosen regarding the existing literature, by takinginto account that the best reconstruction phase goes beyond the margins of this thesis. Both ofthese parameters are multiple of two in order to easily apply the projection over a sparsifyingbasis without loss of coefficients in the specification of the bank of filters which supports thisdomain. The initial aim of the array of electrodes which is related with the acquisitionmultichannel system of this work is the recovery of ECoG and AP signals from epilepsy patients,in order to specify which is the brain area involved with the characteristic seizures theseindividuals sustain. As it has been presented in Table 2.1.1, ECoG and AP signals have abandwidth about 10 kHz, and so the sampling/mixing frequency has to satisfy the Nyquist-Shannon theorem, so a frequency of 30 kHz fits with the sampling requirements of the interestsignals.5.1. Matlab and Simulink ModelsA CS model has been implemented in Simulink in order to simulate the blocks behaviour of theamplification, mixing and integration of the neural signal as it is shown in Fig.3.4.1. Parameterswhich have been integrated in the model are those contained in Table 2.1.1. Every path has47