This example shows how to compute the time-domain response of a simple bandpass filter:

Choose inductance and capacitance values using the classic image parameter design method.

Use

`rfckt.seriesrlc`

,`rfckt.shuntrlc`

and`rfckt.cascade`

to programmatically construct a Butterworth circuit as a 2-port network.Use

`analyze`

to extract the S-parameters of the 2-port network over a wide frequency range.Use

`s2tf`

to compute the voltage transfer function from the input to the output.Use

`rationalfit`

to generate rational fits that capture the ideal RC circuit to a very high degree of accuracy.Create a noisy input voltage waveform.

Use

`timeresp`

to compute the transient response to a noisy input voltage waveform.

The image parameter design method is a framework for analytically computing the values of the series and parallel components in passive filters. For more information on this method, see "Complete Wireless Design" by Cotter W. Sayre, McGraw-Hill 2008 p. 331.

**Figure 1:** A Butterworth bandpass filter built out of two half-sections.

The following MATLAB code generates component values for a bandpass filter with a lower 3-dB cutoff frequency of 2.4 GHz and an upper 3-dB cutoff frequency of 2.5 GHz.

Ro = 50; f1C = 2400e6; f2C = 2500e6; Ls = (Ro / (pi*(f2C - f1C)))/2; Cs = 2*(f2C - f1C)/(4*pi*Ro*f2C*f1C); Lp = 2*Ro*(f2C - f1C)/(4*pi*f2C*f1C); Cp = (1/(pi*Ro*(f2C - f1C)))/2;

The L and C building blocks are formed by selecting appropriate values with the `rfckt.shuntrlc`

function shown in Figure 2 or the `rfckt.seriesrlc`

function shown in Figure 3. The building blocks are then connected together with `rfckt.cascade`

as shown in Figure 4.

**Figure 2:** The 2-port network created by the `rfckt.shuntrlc`

function.

**Figure 3:** The 2-port network created by the `rfckt.seriesrlc`

function.

**Figure 4:** Connecting 2-port networks with the `rfckt.cascade`

function.

Seg1 = rfckt.seriesrlc('L',Ls,'C',Cs); Seg2 = rfckt.shuntrlc('L',Lp,'C',Cp); Seg3 = rfckt.shuntrlc('L',Lp,'C',Cp); Seg4 = rfckt.seriesrlc('L',Ls,'C',Cs); cktBPF = rfckt.cascade('Ckts',{Seg1,Seg2,Seg3,Seg4});

The `analyze`

method extracts the S-parameters from a circuit over a specified vector of frequencies. This example provides a set of frequencies that spans the passband of the filter and analyzes with the default 50-Ohm reference, source impedance, and load impedances. Next, the `s2tf`

function computes the voltage transfer function across the S-parameter model of the circuit. Finally, we generate a high-accuracy rational approximation using the `rationalfit`

function. The resulting approximation matches the network to machine accuracy.

freq = linspace(2e9,3e9,101); analyze(cktBPF,freq); sparams = cktBPF.AnalyzedResult.S_Parameters; tf = s2tf(sparams); fit = rationalfit(freq,tf);

Use the `freqresp`

method to verify that the rational fit approximation has reasonable behavior outside both sides of the fitted frequency range.

widerFreqs = linspace(2e8,5e9,1001); resp = freqresp(fit,widerFreqs); figure semilogy(freq,abs(tf),widerFreqs,abs(resp),'--','LineWidth',2) xlabel('Frequency (Hz)') ylabel('Magnitude') legend('data','fit') title('The rational fit behaves well outside the fitted frequency range.')

This bandpass filter should be able to recover a sinusoidal signal at 2.45 GHz that is made noisy by the inclusion of zero-mean random noise and a blocker at 2.35 GHz. The following MATLAB code constructs such a signal from 4096 samples.

fCenter = 2.45e9; fBlocker = 2.35e9; period = 1/fCenter; sampleTime = period/16; signalLen = 8192; t = (0:signalLen-1)'*sampleTime; % 256 periods input = sin(2*pi*fCenter*t); % Clean input signal rng('default') noise = randn(size(t)) + sin(2*pi*fBlocker*t); noisyInput = input + noise; % Noisy input signal

The `timeresp`

function computes the analytic solution to the state-space equations defined by the rational fit and the input signal.

output = timeresp(fit,noisyInput,sampleTime);

Plot the input signal, noisy input signal, and the band pass filter output in a figure window.

xmax = t(end)/8; figure subplot(3,1,1) plot(t,input) axis([0 xmax -1.5 1.5]) title('Input') subplot(3,1,2) plot(t,noisyInput) axis([0 xmax floor(min(noisyInput)) ceil(max(noisyInput))]) title('Noisy Input') ylabel('Amplitude (volts)') subplot(3,1,3) plot(t,output) axis([0 xmax -1.5 1.5]) title('Filter Output') xlabel('Time (sec)')

Overlaying the noisy input and the filter response in the frequency domain explains why the filtering operation is successful. Both the blocker signal at 2.35 GHz and much of the noise is significantly attenuated.

NFFT = 2^nextpow2(signalLen); % Next power of 2 from length of y Y = fft(noisyInput,NFFT)/signalLen; samplingFreq = 1/sampleTime; f = samplingFreq/2*linspace(0,1,NFFT/2+1)'; O = fft(output,NFFT)/signalLen; figure subplot(2,1,1) plot(freq,abs(tf),'b','LineWidth',2) axis([freq(1) freq(end) 0 1.1]) legend('filter transfer function') ylabel('Magnitude') subplot(2,1,2) plot(f,2*abs(Y(1:NFFT/2+1)),'g',f,2*abs(O(1:NFFT/2+1)),'r','LineWidth',2) axis([freq(1) freq(end) 0 1.1]) legend('input+noise','output') title('Filter characteristic and noisy input spectrum.') xlabel('Frequency (Hz)') ylabel('Magnitude (Volts)')