Communications data

This section presents datasets related to communication systems. The available data are organized into the following categories:

Modem Data

Two sequences of received data from V.29, 9600 bps modems are provided. The first sequence is a short block of training data, of about 1.5 seconds, where the modem goes through a carrier synchronizing procedure, followed by a known PN sequence of symbols. The second sequence is about 4.75 seconds of "random" data. Both sequences are T/2 sampled at 4800 Hz. The carrier has been removed to the limits of practical implementation, but some offset remains. The power level has been normalized to about unity. The data was obtained over a PBX line with a well behaved bandpass response, starting a gradual rollup at about 1100 Hz and a gradual rolloff at about 2300 Hz (the carrier is at about 1700 Hz).

The code provided below is used to load a modem received sequence and to plot its power spectral density. Check Cornell University Blind Equalization Research Group for a demonstration of blind channel equalization of the received data. 

clear;
load data1;

% Plot received signal
figure(1);
plot(data, '.');

% Plot Power Spectral Density (PSD) of the signal
figure(2);
[P, F] = psd(data);
semilogy(F * 1700, fftshift(P));
xlabel('Frequency (Hz)');
ylabel('Power Spectrum Magnitude (dB)');

Cable Data

Two snapshots of received cable channel data streams and two channel models extracted from these original streams are provided. The cable data sequences (data 1 and data 2) have 64K complex-valued samples each, and the channels (channel 1 and channel 2) are modelled by 128 complex-valued taps. Such data are sampled at twice per symbol and the carrier frequency has been mixed close to 0 Hz; note that, there is some residual FM on the carrier.

The code below uses the channel model to find a perfect fractionally spaced equalizer (FSE), which can then be used to equalize the received sequence. The equalizer output eye diagram is plotted in blocks, since the residual FM phase causes a rotation of the constellation in time. 

clear;
load chan2;
HH=convmtx(C,128);
H=HH(1:2:size(HH,1),:);
tau=zeros(size(H,1),1);
tau(64)=1;
f=H\tau;
load data2;
y=conv(f,data);
plot(y(2:2:1000),'.');
title('Eye diagram of equalized cable data');

Microwave Data

This directory contain channel impulse responses (FIR models) obtained from digital microwave radio systems. The complex-valued baseband channel responses have been "fractionally sampled" at twice the baud interval. Such baud intervals vary somewhat from channel to channel, and are listed below.

These responses were extracted from the processing of field measurements of received signals, courtesy of Applied Signal Technology, Sunnyvale, CA.

More information (e.g., frequency response illustrations) can be obtained from Cornell's Blind Equalization Research Group.