I am low pass filtering the PCM of an arbitrary Bell 103 (300 bps) signal. However, the output of the low pass filter shows random spikes in amplitude.

I have tried to increase the bandwidth, but no luck. Can someone explain why these spikes arise? My goal is to do some sort of envelope detection on this signal afterwards, to decode the 1's and 0's. For example by using a rectifier first.

The filter coefficients:

coeffs = [0.265842424474100 0.243276902464240 0.271096721997119 0.243276902464240 0.265842424474100]

The Matlab script used:

close all;
clear all;

% Generate a random bit sequence as input
bitsIn = randi([0 1], 16, 1)';

% Generate the amplitude of the FSK signal
amplitude = randi([7000 10000], 1, 1);

% Generate a randomized length of randomized startup noise
% x = randi([-20 20], 1, randi([0 3000], 1, 1));
x = []; % TODO

% FSK module the input bit stream
for i = 1:length(bitsIn)
    if (bitsIn(i) == 1)
        freq = 1270;
        freq = 1070;

    % Append the new PCM data for the new bit, with random phase
    phase = (2 * rand(1)) * pi;
    x = [x (amplitude * sin((2 * pi * freq * (0:26) / 8000) + phase))];

% Lowpass filter the resulting FSK samples
d = fdesign.lowpass('N,Fp,Fst,Ast',4,1070,1270,10,8000);
Hd = design(d,'equiripple');
y = filter(Hd,x);

title('Input data');
title('Low pass output');
  • $\begingroup$ can you post your filter coefficient ?. This design requires the 'DSP System Toolbox' $\endgroup$ – Hilmar May 19 at 15:31
  • $\begingroup$ @Hilmar Thank you for your response. I have added the coefficients to the OP. $\endgroup$ – LearningDSP May 20 at 6:40

I don't have MatLab with the DSP toolbox, but you're calling filter wrong. Use:

y = filter(Hd, [1], x);

to set the feedback taps properly to $[1]$.

You should also inspect your filter design by using

freqz(Hd, [1], 512, 8000);

Although my hunch is that amplitude spikes in the output are directly caused by the filtering of the phase discontinuities you intentionally added when performing FM modulation.

Really, no modern equipment uses frequency switching to perform FSK. FSK emitted from modern equipment is almost always continuous phase. Are you really dealing with an ancient piece of Bell 103 modem hardware?

An example Octave script that generates the complex I/Q samples of continuous phase FSK is here:


if you need an example of how to generate CPFSK.

| improve this answer | |
  • $\begingroup$ Thank you for your response. I have tried this notation, but it leads to errors. When I check the documentation I did find a notation very similar, but it produced the same results as before. Indeed the freqz() function shows some helpful information. I believe I need the phase discontinuities. I get the signal from 30 year old hardware that conforms with the Bell 103 format. $\endgroup$ – LearningDSP May 20 at 7:00
  • $\begingroup$ The call is correct. Since Hd is an objec he is calling dfilt.dffir/filter() and not the top level filter $\endgroup$ – Hilmar May 21 at 13:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.