tom@455: % Copyright 2012, Google, Inc. dicklyon@462: % Author Richard F. Lyon tom@455: % tom@455: % This Matlab file is part of an implementation of Lyon's cochlear model: tom@455: % "Cascade of Asymmetric Resonators with Fast-Acting Compression" tom@455: % to supplement Lyon's upcoming book "Human and Machine Hearing" tom@455: % tom@455: % Licensed under the Apache License, Version 2.0 (the "License"); tom@455: % you may not use this file except in compliance with the License. tom@455: % You may obtain a copy of the License at tom@455: % tom@455: % http://www.apache.org/licenses/LICENSE-2.0 tom@455: % tom@455: % Unless required by applicable law or agreed to in writing, software tom@455: % distributed under the License is distributed on an "AS IS" BASIS, tom@455: % WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. tom@455: % See the License for the specific language governing permissions and tom@455: % limitations under the License. tom@455: tom@455: function [naps, CF, decim_naps] = CARFAC_Run ... tom@455: (CF, input_waves, AGC_plot_fig_num) dicklyon@462: % function [naps, CF, decim_naps] = CARFAC_Run ... dicklyon@462: % (CF, input_waves, AGC_plot_fig_num) tom@455: % This function runs the CARFAC; that is, filters a 1 or more channel tom@455: % sound input to make one or more neural activity patterns (naps). tom@455: % tom@455: % The CF struct holds the filterbank design and state; if you want to tom@455: % break the input up into segments, you need to use the updated CF tom@455: % to keep the state between segments. tom@455: % tom@455: % input_waves is a column vector if there's just one audio channel; tom@455: % more generally, it has a row per time sample, a column per audio channel. tom@455: % tom@455: % naps has a row per time sample, a column per filterbank channel, and tom@455: % a layer per audio channel if more than 1. tom@455: % decim_naps is like naps but time-decimated by the int CF.decimation. tom@455: % tom@455: % the input_waves are assumed to be sampled at the same rate as the tom@455: % CARFAC is designed for; a resampling may be needed before calling this. tom@455: % tom@455: % The function works as an outer iteration on time, updating all the tom@455: % filters and AGC states concurrently, so that the different channels can tom@455: % interact easily. The inner loops are over filterbank channels, and tom@455: % this level should be kept efficient. tom@455: % tom@455: % See other functions for designing and characterizing the CARFAC: tom@455: % CF = CARFAC_Design(fs, CF_filter_params, CF_AGC_params, n_mics) tom@455: % transfns = CARFAC_Transfer_Functions(CF, to_chans, from_chans) tom@455: tom@455: [n_samp, n_mics] = size(input_waves); tom@455: n_ch = CF.n_ch; tom@455: tom@455: if nargin < 3 tom@455: AGC_plot_fig_num = 0; tom@455: end tom@455: tom@455: if n_mics ~= CF.n_mics tom@455: error('bad number of input_waves channels passed to CARFAC_Run') tom@455: end tom@455: dicklyon@462: % fastest decimated rate determines some interp needed: dicklyon@462: decim1 = CF.AGC_params.decimation(1); tom@455: tom@455: naps = zeros(n_samp, n_ch, n_mics); dicklyon@462: decim_k = 0; dicklyon@462: k_NAP_decim = 0; dicklyon@462: NAP_decim = 8; tom@455: if nargout > 2 tom@455: % make decimated detect output: dicklyon@462: decim_naps = zeros(ceil(n_samp/NAP_decim), CF.n_ch, CF.n_mics); tom@455: else tom@455: decim_naps = []; tom@455: end tom@455: tom@455: dicklyon@462: k_AGC = 0; dicklyon@462: AGC_plot_decim = 16; % how often to plot AGC state; TODO: use segments tom@455: dicklyon@462: dicklyon@462: detects = zeros(n_ch, n_mics); tom@455: for k = 1:n_samp dicklyon@462: CF.k_mod_decim = mod(CF.k_mod_decim + 1, decim1); % global time phase dicklyon@462: k_NAP_decim = mod(k_NAP_decim + 1, NAP_decim); % phase of decimated nap tom@455: % at each time step, possibly handle multiple channels tom@455: for mic = 1:n_mics tom@455: [filters_out, CF.filter_state(mic)] = CARFAC_FilterStep( ... tom@455: input_waves(k, mic), CF.filter_coeffs, CF.filter_state(mic)); dicklyon@462: tom@455: % update IHC state & output on every time step, too tom@455: [ihc_out, CF.IHC_state(mic)] = CARFAC_IHCStep( ... tom@455: filters_out, CF.IHC_coeffs, CF.IHC_state(mic)); dicklyon@462: dicklyon@462: detects(:, mic) = ihc_out; % for input to AGC, and out to SAI dicklyon@462: tom@455: naps(k, :, mic) = ihc_out; % output to neural activity pattern dicklyon@462: tom@455: end dicklyon@462: if ~isempty(decim_naps) && (k_NAP_decim == 0) dicklyon@462: decim_k = decim_k + 1; % index of decimated NAP dicklyon@462: for mic = 1:n_mics dicklyon@462: decim_naps(decim_k, :, mic) = CF.IHC_state(mic).ihc_accum / ... dicklyon@462: NAP_decim; % for cochleagram dicklyon@462: CF.IHC_state(mic).ihc_accum = zeros(n_ch,1); tom@455: end dicklyon@462: end dicklyon@462: % run the AGC update step, taking input from IHC_state, decimating dicklyon@462: % internally, all mics at once due to mixing across them: dicklyon@462: [CF.AGC_state, updated] = ... dicklyon@462: CARFAC_AGCStep(CF.AGC_coeffs, detects, CF.AGC_state); dicklyon@462: dicklyon@462: % connect the feedback from AGC_state to filter_state when it updates dicklyon@462: if updated tom@455: for mic = 1:n_mics dicklyon@462: new_damping = CF.AGC_state(mic).AGC_memory(:, 1); % stage 1 result tom@455: % set the delta needed to get to new_damping: dicklyon@462: % TODO: update this to use da and dc instead of dr maybe? tom@455: CF.filter_state(mic).dzB_memory = ... tom@455: (new_damping - CF.filter_state(mic).zB_memory) ... dicklyon@462: / decim1; tom@455: end tom@455: end dicklyon@462: dicklyon@462: k_AGC = mod(k_AGC + 1, AGC_plot_decim); dicklyon@462: if AGC_plot_fig_num && k_AGC == 0 dicklyon@462: figure(AGC_plot_fig_num); hold off; clf dicklyon@462: set(gca, 'Position', [.25, .25, .5, .5]) dicklyon@462: dicklyon@462: maxsum = 0; dicklyon@462: for mic = 1:n_mics dicklyon@462: plot(CF.AGC_state(mic).AGC_memory(:, 1), 'k-', 'LineWidth', 1) dicklyon@462: maxes(mic) = max(CF.AGC_state(mic).AGC_memory(:)); dicklyon@462: hold on dicklyon@462: stage1 = 4; % as opposed to stage dicklyon@462: for stage = 1:3; dicklyon@462: plot(2^(stage1-1) * (CF.AGC_state(mic).AGC_memory(:, stage) - ... dicklyon@462: 2 * CF.AGC_state(mic).AGC_memory(:, stage+1))); dicklyon@462: end dicklyon@462: stage = 4; dicklyon@462: plot(2^(stage1-1) * CF.AGC_state(mic).AGC_memory(:, stage)); dicklyon@462: end dicklyon@462: axis([0, CF.n_ch+1, -0.01, max(maxes) + 0.01]); dicklyon@462: drawnow dicklyon@462: end dicklyon@462: tom@455: end tom@455: