% minimal example IM/DD M = 4; fsym = 180e9; apply_pulsef = 1; fdac = 256e9; fadc = 256e9; random_key = 1; rcalpha = 0.05; kover = 16; duob_mode = db_mode.no_db; vbias_rel = 0.5; u_pi = 3; vbias = -vbias_rel*u_pi; laser_wavelength = 1293; laser_linewidth = 0; tx_bw_nyquist = 0.8; % Channel link_length = 1; % RX rop = -7; rx_bw_nyquist = 0.8; vnle_order1 = 50; vnle_order2 = 7; vnle_order3 = 7; vnle_order=[vnle_order1,vnle_order2,vnle_order3]; dfe_order = [0 0 0]; pf_ncoeffs = 1; alpha = 0; len_tr = 4096*2; mu_ffe1 = 0.0001; mu_ffe2 = 0.0008; mu_ffe3 = 0.001; mu_dc = 0.005; % mu_dc = 0; mu_ffe = [mu_ffe1 mu_ffe3 mu_ffe3]; mu_dfe = 0.0004; n_stat_runs = 1; base_random_key = random_key; output.eq_stats = struct(); for stat_run = 1:n_stat_runs random_key = base_random_key + stat_run - 1; Pform = Pulseformer("fsym",fsym,"fdac",4*fsym,"pulse","rrc","pulselength",16,"alpha",rcalpha); [Digi_sig,Symbols,Tx_bits] = PAMsource(... "fsym",fsym,"M",M,"order",16,"useprbs",0,... "fs_out",fdac,... "applyclipping",0,"clipfactor",1.5,... "applypulseform",apply_pulsef,"pulseformer",Pform,... "randkey",random_key,... 'duobinary_mode',duob_mode,... "mrds_code",0,"mrds_blocklength",512).process(); %%%%% AWG El_sig = M8199A("kover",kover).process(Digi_sig); % El_sig = AWG("fdac",fdac,"f_cutoff",fsym,"lpf_active",0,"kover",kover,"bit_resolution",12,"upsampling_method","samplehold","precomp_sinc_rolloff",1).process(Digi_sig); El_sig.spectrum("displayname",'Digi Spectrum','fignum',1,'normalizeTo0dB',1); xlim([0,130]); ylim([-30,5]); % El_sig = El_sig.setPower(0,"dBm"); %%%%% Electrical Driver Amplifier %%%%%% % El_sig = Amplifier("amp_mode","ideal_no_noise","gain_mode","gain","amplification_db",3).process(El_sig); El_sig = El_sig.normalize("mode","oneone"); scaling = 0.6*(u_pi/2-abs(vbias-u_pi/2)); El_sig = El_sig .* scaling; %%%%% MODULATE E/O CONVERSION %%%%%% [Opt_sig] = EML("mode",eml_mode.im_cosinus,"power",3,"fsimu",El_sig.fs,"lambda",laser_wavelength,"bias",vbias,"u_pi",u_pi,"linewidth",laser_linewidth,"randomkey",random_key+1,"alpha",alpha).process(El_sig); Opt_sig.spectrum("displayname",'Opt Spectrum','fignum',10,'normalizeTo0dB',1); % Opt_sig.eye(fsym,M,"displayname",'eye adter modulator','fignum',2026); Opt_sig = Fiber("fsimu",Opt_sig.fs,"fiber_length",link_length,"alpha",0.3,"D",0,"lambda0",1310,"gamma",0,"Dslope",0.07).process(Opt_sig); %%%%%% ROP %%%%%% Rx_sig = Amplifier("amp_mode","ideal_no_noise","gain_mode","output_power","amplification_db",rop).process(Opt_sig); %%%%%% PD Square Law %%%%%% Rx_sig = Photodiode("fsimu",fdac*kover,"dark_current",2e-08,"responsivity",1,"temperature",20,"nep",1.8e-11).process(Rx_sig); %%%%%% Low-pass RX (PD, El. Connectors and Scope %%%%%% rx_bwl = 80e9; Rx_sig = Filter('filtdegree',4,"f_cutoff",rx_bwl,"fs",fdac*kover,"filterType",filtertypes.butterworth,"active",true).process(Rx_sig); % %%%%%% Low-pass Scope %%%%%% Lp_scpe = Filter('filtdegree',4,"f_cutoff",110e9,"fs",fadc,"filterType",filtertypes.butterworth,"active",true); %%%%%% Scope %%%%%% Scpe_sig = Scope("fsimu",fdac*kover,"fadc",fadc,... "delay",0,"fixed_delay",0,"filtertype",filtertypes.butterworth,... "samplingdelay",0,"rand_samplingdelay",0,"freq_offset",0,"samp_jitter",0,... "adcresolution",8,"quantbuffer",0.1,'block_dc',1,'lpf_active',1,'H_lpf',Lp_scpe).process(Rx_sig); %% % 1) matched filter % pulse is symmetric, hence we can use pulsef firectly as matched filter. % It feels off (bit I think correct) that the fsym is now the output freq.!! % -> output 2 sps to omit timing recovery!? Pform = Pulseformer("fsym",fsym,"fdac",2*fsym,"pulse","rrc","pulselength",16,"alpha",rcalpha,"matched",1); Scpe_sig = Pform.process(Scpe_sig); Scpe_sig.spectrum("displayname",'Signal after matched filter','fignum',1,'normalizeTo0dB',1); % % %% % %%%%%% Sample to 2x fsym %%%%%% % Scpe_sig = Scpe_sig.resample("fs_out",2*fsym); % Scpe_sig.signal = Scpe_sig.signal(1:2*length(Symbols)); %% %%%%%% Sync Rx signal with reference %%%%%% [Scpe_sig,~] = Scpe_sig.tsynch("reference",Symbols,"fs_ref",fsym,"debug_plots",0); Scpe_sig.spectrum("displayname",'Opt Spectrum','fignum',11,'normalizeTo0dB',1); % Scpe_sig = Filter('filtdegree',4,"f_cutoff",Symbols.fs.*0.5,"fs",Scpe_sig.fs,"filterType",filtertypes.gaussian,"active",true).process(Scpe_sig); Scpe_sig = Scpe_sig - mean(Scpe_sig.signal); Scpe_sig.signal = Scpe_sig.signal(1:2*length(Symbols)); %% -------------------- FFE -------------------- ffe_options = struct( ... "epochs_tr", 3, ... "epochs_dd", 3, ... "len_tr", 4096, ... "order", 40, ... "sps", 2, ... "decide", 0, ... "adaption", adaption_method.rls, ... "dd_mode", 1); eq_nlms = FFE("epochs_tr",ffe_options.epochs_tr,"epochs_dd",ffe_options.epochs_dd, ... "len_tr",ffe_options.len_tr,"mu_dd",0.001,"mu_tr",0.001, ... "order",ffe_options.order,"sps",ffe_options.sps,"decide",ffe_options.decide, ... "adaption",ffe_options.adaption,"dd_mode",ffe_options.dd_mode,"save_debug",1,"optmize_mus",1); output.ffe_results = ffe(eq_nlms,M,Scpe_sig,Symbols,Tx_bits, ... "precode_mode",duob_mode,'showAnalysis',0,"postFFE",[], ... "eth_style_symbol_mapping",0); output.ffe_results.metrics.print("description","FFE"); %% % -------------------- FFE -------------------- %% for i = 3:-1:1 mu_adaptation = adaption_method(i); switch mu_adaptation case adaption_method.lms mu_search_range = [1e-5, 1e-2]; case adaption_method.nlms mu_search_range = [1e-3, 5e-1]; case adaption_method.rls mu_search_range = [0.98, 0.99999]; % here mu is lambda end mu_max_evaluations = 10; ffe_options = struct( ... "epochs_tr", 3, ... "epochs_dd", 2, ... "len_tr", 4096*2, ... "order", 20, ... "sps", 2, ... "decide", 0, ... "adaption", mu_adaptation, ... "dd_mode", 1); method_name = char(mu_adaptation); rng(random_key, "twister"); mu_dd_var = optimizableVariable("mu_dd", mu_search_range, "Transform", "log"); mu_tr_var = optimizableVariable("mu_tr", mu_search_range, "Transform", "log"); global ffe_mu_trace ffe_mu_trace = table('Size',[0 5], ... 'VariableTypes',{'double','double','double','double','double'}, ... 'VariableNames',{'mu_dd','mu_tr','BER','SNR','mse_last_epoch'}); objective_fn = @(params) ffe_mu_objective(params, ffe_options, M, Scpe_sig, Symbols, Tx_bits, duob_mode); output.eq_stats.(method_name).mu_optimization{stat_run} = bayesopt(objective_fn, [mu_dd_var, mu_tr_var], ... "MaxObjectiveEvaluations", mu_max_evaluations, ... "AcquisitionFunctionName", "expected-improvement-plus", ... "IsObjectiveDeterministic", false, ... "Verbose", 1, ... "PlotFcn", []); output.eq_stats.(method_name).mu_observed(stat_run,:) = output.eq_stats.(method_name).mu_optimization{stat_run}.XAtMinObjective; output.eq_stats.(method_name).mu_estimated(stat_run,:) = output.eq_stats.(method_name).mu_optimization{stat_run}.XAtMinEstimatedObjective; output.eq_stats.(method_name).optimization_trace{stat_run} = ffe_mu_trace; best_mu_dd = output.eq_stats.(method_name).mu_observed.mu_dd(stat_run); best_mu_tr = output.eq_stats.(method_name).mu_observed.mu_tr(stat_run); output.eq_stats.(method_name).mu_used(stat_run,:) = table(best_mu_dd,best_mu_tr, ... 'VariableNames',{'mu_dd','mu_tr'}); eq_nlms = FFE("epochs_tr",ffe_options.epochs_tr,"epochs_dd",ffe_options.epochs_dd, ... "len_tr",ffe_options.len_tr,"mu_dd",best_mu_dd,"mu_tr",best_mu_tr, ... "order",ffe_options.order,"sps",ffe_options.sps,"decide",ffe_options.decide, ... "adaption",ffe_options.adaption,"dd_mode",ffe_options.dd_mode,"save_debug",1); output.ffe_results = ffe(eq_nlms,M,Scpe_sig,Symbols,Tx_bits, ... "precode_mode",duob_mode,'showAnalysis',0,"postFFE",[], ... "eth_style_symbol_mapping",0); output.ffe_results.metrics.print("description",sprintf('%s',mu_adaptation)); output.eq_stats.(method_name).error_tr_first_epoch(stat_run,:) = eq_nlms.debug_struct.error_tr(1,:); % figure(6) % hold on % for epoch = 1:ffe_options.epochs_tr % plot(1:size(eq_nlms.debug_struct.error_tr,2),movmean(eq_nlms.debug_struct.error_tr(epoch,:),10)); % end end end %% figure(5); clf; hold on; method_names = fieldnames(output.eq_stats); cols = cbrewer2('Paired',6); cols = cols(2:2:end,:); for i = 1:numel(method_names) method_name = method_names{i}; first_epoch = output.eq_stats.(method_name).error_tr_first_epoch; for j = 1:size(first_epoch,1) plot(movmean(output.eq_stats.(method_name).error_tr_first_epoch(j,:),10), ... "DisplayName",method_name,"Color",[0.7,0.7,0.7],"LineWidth",0.1,'HandleVisibility','off'); end end for i = 1:numel(method_names) method_name = method_names{i}; first_epoch = output.eq_stats.(method_name).error_tr_first_epoch; output.eq_stats.(method_name).error_tr_first_epoch_mean = mean(first_epoch,1,"omitnan"); output.eq_stats.(method_name).error_tr_first_epoch_std = std(first_epoch,0,1,"omitnan"); plot(movmean(output.eq_stats.(method_name).error_tr_first_epoch_mean,50), ... "DisplayName",method_name,"Color",cols(i,:)); end grid on; xlabel("Training symbol index","Interpreter","tex"); ylabel("Mean squared error, first epoch","Interpreter","tex"); legend("Location","best","Interpreter","tex"); figure(6); clf; tiledlayout(numel(method_names),3,"TileSpacing","compact","Padding","compact"); metric_names = {'BER','SNR','mse_last_epoch'}; metric_labels = {'BER','SNR','mean abs(error)^2, last epoch'}; for i = 1:numel(method_names) method_name = method_names{i}; trace = vertcat(output.eq_stats.(method_name).optimization_trace{:}); observed_mu = output.eq_stats.(method_name).mu_observed; estimated_mu = output.eq_stats.(method_name).mu_estimated; for metric_idx = 1:numel(metric_names) nexttile; metric = trace.(metric_names{metric_idx}); if metric_idx ~= 2 metric = log10(max(metric,realmin)); end scatter(trace.mu_tr,trace.mu_dd,25,metric,"filled"); hold on; plot(observed_mu.mu_tr,observed_mu.mu_dd,"kp","MarkerSize",11, ... "MarkerFaceColor","y","LineWidth",1.2,"DisplayName","Observed best"); plot(estimated_mu.mu_tr,estimated_mu.mu_dd,"kx","MarkerSize",10, ... "LineWidth",1.6,"DisplayName","Estimated best"); set(gca,"XScale","log","YScale","log"); grid on; xlabel('\mu_{tr}',"Interpreter","tex"); ylabel('\mu_{dd}',"Interpreter","tex"); title(sprintf("%s: %s",method_name,metric_labels{metric_idx}),"Interpreter","tex"); cb = colorbar; if metric_idx ~= 2 cb.Label.String = sprintf("log10(%s)",metric_labels{metric_idx}); else cb.Label.String = metric_labels{metric_idx}; end cb.Label.Interpreter = "tex"; legend("Location","best","Interpreter","tex"); end end function objective = ffe_mu_objective(params, ffe_options, M, rx_signal, tx_symbols, tx_bits, duob_mode) global ffe_mu_trace try eq_candidate = FFE("epochs_tr",ffe_options.epochs_tr,"epochs_dd",ffe_options.epochs_dd, ... "len_tr",ffe_options.len_tr,"mu_dd",params.mu_dd,"mu_tr",params.mu_tr, ... "order",ffe_options.order,"sps",ffe_options.sps,"decide",ffe_options.decide, ... "adaption",ffe_options.adaption,"dd_mode",ffe_options.dd_mode,"save_debug",1); candidate_results = ffe(eq_candidate, M, rx_signal, tx_symbols, tx_bits, ... "precode_mode",duob_mode,'showAnalysis',0,"postFFE",[], ... "eth_style_symbol_mapping",0); objective = candidate_results.metrics.BER; if ~isfinite(objective) objective = 0.5; end mse_last_epoch = mean(eq_candidate.debug_struct.error_tr(end,:),"omitnan"); ffe_mu_trace = [ffe_mu_trace; table(params.mu_dd,params.mu_tr,objective, ... candidate_results.metrics.SNR,mse_last_epoch, ... 'VariableNames',{'mu_dd','mu_tr','BER','SNR','mse_last_epoch'})]; fprintf("mu_dd=%9.3e, mu_tr=%9.3e -> BER=%9.3e, SNR=%6.2f dB, MSE=%9.3e\n", ... params.mu_dd, params.mu_tr, objective, candidate_results.metrics.SNR, mse_last_epoch); catch ME objective = 0.5; ffe_mu_trace = [ffe_mu_trace; table(params.mu_dd,params.mu_tr,objective,NaN,NaN, ... 'VariableNames',{'mu_dd','mu_tr','BER','SNR','mse_last_epoch'})]; fprintf("mu_dd=%9.3e, mu_tr=%9.3e -> failed: %s\n", ... params.mu_dd, params.mu_tr, ME.message); end end