Capture and Label NR and LTE Signals for AI Training

Since R2023b

This example shows how to scan, capture, and label bandwidths with 5G NR and LTE signals using an NI™ USRP™ software-defined radio (SDR). The example scans a wide bandwidth to determine 5G NR and LTE primary synchronization signal (PSS) locations and captures and interactively labels the bandwidth of the associated carrier waveforms and any unknown signals.

Associated AI for Wireless Examples

Use this example as part of a deep learning workflow:

1. Use this example to scan, capture, and label bandwidths with 5G NR and LTE signals using an SDR.

2. Use the Spectrum Sensing with Deep Learning to Identify 5G, LTE, and WLAN Signals example to train a semantic segmentation network to identify 5G NR and LTE signals in a wideband spectrogram.

3. Use the Identify LTE and NR Signals from Captured Data Using SDR and Deep Learning example to use a deep learning trained semantic segmentation network to identify NR and LTE signals from wireless data captured with an SDR.

This diagram shows the complete workflow.

Introduction

In this example, you use hNRPSSDescriptor and hLTEPSSDescriptor helper objects to describe 5G NR and LTE signal PSS signals that you want to detect and capture. Using the configureDetector helper function, you then configure an SDR preamble detector to scan for each signal by setting the following parameters:

• Center frequency

• Sample rate

• Preamble

• Trigger Offset

After you detect the signals, you capture the full bandwidth and postprocess the data by labeling each associated carrier waveform. This diagram shows the workflow.

Set Up Radio

Call the radioConfigurations function. The function returns all available radio setup configurations that you saved using the Radio Setup wizard.

savedRadioConfigurations = radioConfigurations;

To update the dropdown menu with your saved radio setup configuration names, click Update. Then select the radio to use with this example.

savedRadioConfigurationNames = [string({savedRadioConfigurations.Name})];
radio = savedRadioConfigurationNames(1) ;

Choose Bandwidths to Capture and Label

Set the center frequencies and bandwidth of the captures. Use the hMaxBandwidth helper function to set the bandwidth to the maximum bandwidth of your radio.

centerFrequencies = [2600e6 3550e6];
maxBandwidth = hMaxBandwidth(radio);

Describe Wireless Signals

Describe the 5G NR and LTE PSS that might be present in each bandwidth. Use the hNRPSSDescriptorsInFrequencyBand and hLTEPSSDescriptorsInFrequencyBand helper functions to create hNRPSSDescriptor and hLTEPSSDescriptor helper objects, respectively, according to the expected signal locations in the standard.

captureBandwidths = [];
numPotentialSignals = 0;
for cf = centerFrequencies
    wsds = [hNRPSSDescriptorsInFrequencyBand(cf-maxBandwidth/2,cf+maxBandwidth/2), ...
        hLTEPSSDescriptorsInFrequencyBand(cf-maxBandwidth/2,cf+maxBandwidth/2)];
    captureBandwidths = [captureBandwidths, struct(CenterFrequency=cf,Bandwidth=maxBandwidth,SignalDescriptors=wsds)];
    numPotentialSignals = numPotentialSignals+length(wsds);
end

Configure and Tune Radio as Preamble Detector

Configure Preamble Detector

Create a preambleDetector object with the specified radio. Because the object requires exclusive access to radio hardware resources, clear any other object associated with the specified radio.

if ~exist("pd","var")
    clear bbrx
    pd = preambleDetector(radio);
end

To update the dropdown menu with the antennas available for capture on your radio, call the hCaptureAntennas helper function. Then select the antenna to use with this example.

captureAntennaSelection = hCaptureAntennas(radio);
pd.Antennas = captureAntennaSelection(1);

Tune Preamble Detector

Set the adaptive threshold gain, the adaptive threshold offset, and the radio gain values of the preamble detector for the local environment. You can calibrate these values by exploring the trigger points that the plotThreshold function provides. For more information, see Triggered Capture Using Preamble Detection.

pd.ThresholdMethod = "adaptive";
pd.AdaptiveThresholdGain = 0.3;
pd.AdaptiveThresholdOffset = 0.1;
pd.RadioGain = 30;

Scan for Wireless Signals

Use the configureDetector helper function to configure the preamble detector with an appropriate preamble, trigger offset, sample rate and center frequency. Set the detection timeout to 20 milliseconds, which is the expected periodicity of a 5G NR SSB. Then use the capture function to detect and capture the described NR or LTE signal. If the detection is successful, save the detected signal descriptor. For large bands, the scan takes some time.

wb = hCaptureWaitBar(numPotentialSignals);
for lb = 1:length(captureBandwidths)
    detectedSignals = [];
    for wsd = captureBandwidths(lb).SignalDescriptors
        hUpdateCaptureWaitBar(wb,wsd)
        pd = configureDetector(wsd,pd);
        [~,~,~,status] = capture(pd,1, milliseconds(20));
        if status
            % Signal detected
            detectedSignals = [detectedSignals,wsd];
            disp("Detected "+string(class(wsd))+" at "+string(wsd.CenterFrequency/1e6)+" MHz")
        end
    end
    % Update labeled bandwidth structure with only detected signals
    captureBandwidths(lb).SignalDescriptors=detectedSignals;
end

Loading bitstream to FPGA...
Loading bitstream to FPGA is now complete.

Detected hLTEPSSDescriptor at 806 MHz
Detected hLTEPSSDescriptor at 805.7 MHz
Detected hLTEPSSDescriptor at 805.8 MHz
Detected hLTEPSSDescriptor at 805.9 MHz
Detected hLTEPSSDescriptor at 806 MHz
Detected hLTEPSSDescriptor at 806.1 MHz
Detected hLTEPSSDescriptor at 806.2 MHz
Detected hLTEPSSDescriptor at 806.3 MHz
Detected hLTEPSSDescriptor at 816 MHz
Detected hNRPSSDescriptor at 2647.45 MHz
Detected hLTEPSSDescriptor at 2662.9 MHz
Detected hLTEPSSDescriptor at 2679.7 MHz
Detected hLTEPSSDescriptor at 2679.8 MHz
Detected hLTEPSSDescriptor at 2679.9 MHz
Detected hLTEPSSDescriptor at 2680 MHz
Detected hLTEPSSDescriptor at 2680.1 MHz
Detected hLTEPSSDescriptor at 2680.2 MHz
Detected hLTEPSSDescriptor at 2680.3 MHz
Detected hLTEPSSDescriptor at 2629.7 MHz
Detected hLTEPSSDescriptor at 2630 MHz
Detected hNRPSSDescriptor at 2159.05 MHz
Detected hLTEPSSDescriptor at 2161.6 MHz
Detected hLTEPSSDescriptor at 2161.9 MHz
Detected hLTEPSSDescriptor at 2162 MHz
Detected hLTEPSSDescriptor at 2162.1 MHz
Detected hLTEPSSDescriptor at 2162.2 MHz
Detected hLTEPSSDescriptor at 2162.3 MHz
Detected hLTEPSSDescriptor at 2162.4 MHz
Detected hLTEPSSDescriptor at 2162.5 MHz
Detected hLTEPSSDescriptor at 2119.5 MHz
Detected hLTEPSSDescriptor at 2119.6 MHz
Detected hLTEPSSDescriptor at 2119.7 MHz
Detected hLTEPSSDescriptor at 2119.8 MHz
Detected hLTEPSSDescriptor at 2120 MHz
Detected hLTEPSSDescriptor at 2120.1 MHz
Detected hLTEPSSDescriptor at 2129.6 MHz
Detected hLTEPSSDescriptor at 2129.7 MHz
Detected hLTEPSSDescriptor at 2129.9 MHz
Detected hLTEPSSDescriptor at 1807.6 MHz
Detected hLTEPSSDescriptor at 1807.9 MHz
Detected hLTEPSSDescriptor at 1846.1 MHz
Detected hLTEPSSDescriptor at 1846.1 MHz
Detected hLTEPSSDescriptor at 1846.4 MHz
Detected hLTEPSSDescriptor at 1846.5 MHz
Detected hLTEPSSDescriptor at 1846.6 MHz
Detected hLTEPSSDescriptor at 1846.6 MHz
Detected hLTEPSSDescriptor at 1846.7 MHz
Detected hLTEPSSDescriptor at 1846.8 MHz
Detected hLTEPSSDescriptor at 1846.8 MHz
Detected hLTEPSSDescriptor at 1846.9 MHz
Detected hLTEPSSDescriptor at 1847 MHz
Detected hLTEPSSDescriptor at 1847.3 MHz
Detected hLTEPSSDescriptor at 1865.9 MHz
Detected hLTEPSSDescriptor at 1866.2 MHz
Detected hLTEPSSDescriptor at 1866.3 MHz
Detected hLTEPSSDescriptor at 1866.3 MHz
Detected hLTEPSSDescriptor at 1866.4 MHz
Detected hLTEPSSDescriptor at 1866.5 MHz
Detected hLTEPSSDescriptor at 1866.6 MHz
Detected hLTEPSSDescriptor at 1866.7 MHz
Detected hLTEPSSDescriptor at 1866.8 MHz
Detected hLTEPSSDescriptor at 1866.8 MHz
Detected hLTEPSSDescriptor at 1807.6 MHz
Detected hLTEPSSDescriptor at 1846.7 MHz
Detected hLTEPSSDescriptor at 1860.8 MHz
Detected hLTEPSSDescriptor at 1861 MHz
Detected hLTEPSSDescriptor at 1861.1 MHz
Detected hLTEPSSDescriptor at 1861.1 MHz
Detected hLTEPSSDescriptor at 1861.2 MHz
Detected hLTEPSSDescriptor at 1861.3 MHz
Detected hLTEPSSDescriptor at 1861.3 MHz
Detected hLTEPSSDescriptor at 1861.4 MHz
Detected hLTEPSSDescriptor at 942.4 MHz
Detected hLTEPSSDescriptor at 952.5 MHz

close(wb)

Configure Radio as Baseband Receiver

Configure Baseband Receiver

Create a basebandReceiver object with the specified radio. Clear any other object associated with the specified radio.

if ~exist("bbrx","var")
    clear pd
    bbrx = basebandReceiver(radio);
end

To update the dropdown menu with the antennas available for capture on your radio, call the hCaptureAntennas helper function. Then select the number of antennas to capture with. Use the dropdown menus to select the individual antennas.

captureAntennaSelection = hCaptureAntennas(radio);
numAntennas = string(1:length(captureAntennaSelection));
numCaptureAntennas = numAntennas(1); % Number of antennas to capture
selectedAntennas = captureAntennaSelection(1); % First antenna
if str2double(numCaptureAntennas)>=2
    selectedAntennas(2) = captureAntennaSelection(1); % Second antenna
end
if str2double(numCaptureAntennas)>=3
    selectedAntennas(3) = captureAntennaSelection(1); % Third antenna
end
if str2double(numCaptureAntennas)>=4
    selectedAntennas(4) = captureAntennaSelection(1); % Fourth antenna
end
bbrx.Antennas = selectedAntennas;

Tune Baseband Receiver

Set the radio gain value of the baseband receiver for the local environment. For more information, see Calibrate Radio Gain for Signal Capture.

bbrx.RadioGain = 30;

Capture Bandwidths to Label

Configure the baseband receiver with an appropriate sample rate and center frequency. Set the sample rate to a value greater than the bandwidth. Use the hMaxSampleRate helper function to set the sample rate to the maximum sample rate of your radio. Then use the capture function to capture 200 milliseconds of data, which corresponds to five frames of training data in the Spectrum Sensing with Deep Learning to Identify 5G, LTE, and WLAN Signals example.

bbrx.SampleRate=hMaxSampleRate(radio);
for lb = 1:length(captureBandwidths)
    bbrx.CenterFrequency=captureBandwidths(lb).CenterFrequency;
    [captureBandwidths(lb).Data,captureBandwidths(lb).Timestamp] = capture(bbrx, milliseconds(200));
    captureBandwidths(lb).SampleRate = bbrx.SampleRate;
end

Loading bitstream to FPGA...
Loading bitstream to FPGA is now complete.

Store Captured Data

Store the captured data in MAT files.

parentDir = tempdir;
dataDir = fullfile(parentDir, "capturedData");
if ~exist(dataDir, "dir")
    mkdir(dataDir);
end

for lb = 1:length(captureBandwidths)
    % Save to file
    labeledBandwidth = captureBandwidths(lb);
    disp(labeledBandwidth);

    % Resample data into chunks
    dataLength = min(100e6,length(labeledBandwidth.Data(:,1)));
    bw = labeledBandwidth.Bandwidth;
    sr = labeledBandwidth.SampleRate;
    filteredData = lowpass(single(labeledBandwidth.Data(1:dataLength,1)),bw/2,sr);
    resampledData = resample(filteredData,bw,sr);
    sampleRate = bw;
    resampledDataTotal = buffer(resampledData, 1e7);

    for sdi = 1:size(resampledDataTotal,2)
        resampledData = resampledDataTotal(:,sdi);
        fname = "CF"+strrep(string(labeledBandwidth.CenterFrequency/1e6),".",",")+ ...
            "MHz_BW"+string(labeledBandwidth.Bandwidth/1e6)+"MHz_"+ ...
            string(datetime(labeledBandwidth.Timestamp,"Format","yyyyMMddHHmmssSSS"))+"_"+sdi;
        fname = fullfile(dataDir, fname);
        save(fname, "labeledBandwidth","resampledData","sampleRate","-v7.3")
    end
end

      CenterFrequency: 800000000
            Bandwidth: 200000000
    SignalDescriptors: [1×5 hLTEPSSDescriptor]
                 Data: [50000000×1 int16]
            Timestamp: 21-Jan-2026 18:59:19
           SampleRate: 250000000

      CenterFrequency: 2.6000e+09
            Bandwidth: 200000000
    SignalDescriptors: [1×26 wt.internal.WirelessSignalDescriptor]
                 Data: [50000000×1 int16]
            Timestamp: 21-Jan-2026 18:59:23
           SampleRate: 250000000

      CenterFrequency: 3.5500e+09
            Bandwidth: 200000000
    SignalDescriptors: [1×1 hNRPSSDescriptor]
                 Data: [50000000×1 int16]
            Timestamp: 21-Jan-2026 18:59:26
           SampleRate: 250000000

Display Captured Data and Label Signal Bandwidth

Use the Signal Labeler app to plot a spectrogram and interactively label the data from the first antenna at each bandwidth.

Create a labeled signal set object for the Signal Labeler app to import. The labeledSignalSet object contains the time and frequency limits of each label detected, as well as the corresponding label value.

% Define spectrogram computation
tfOpts = labelSpectrogramOptions("windowlength", ...
    WindowLengthMode="specify",WindowLength=256,Window="hann", ...
    NFFTMode="specify",NFFT=4096,UseDecibels=true, ...
    Overlap=(10/256)*100,MinimumThresholdMode="specify", ...
    MinimumThreshold=-90);

% Set up label definition for specifying label details
labelDef = signalLabelDefinition("SpectrumSensing", ...
    LabelDataType="categorical", ...
    Categories=["LTE" "NR" "Unknown" "None"], ...
    LabelType="roiTimeFrequency", ...
    TimeFrequencyOptions=tfOpts);

% Create a labeledSignalSet object from files
sds = signalDatastore(dataDir,SignalVariableNames="resampledData",SampleRateVariableName="sampleRate");
lss = labeledSignalSet(sds);
lss.addLabelDefinitions(labelDef);

% Add preliminary labels that were captured
resampledFiles = strsplit(strtrim(string(ls(fullfile(dataDir,"*.mat")))));

for fdx = 1:length(resampledFiles)
    fileName = resampledFiles(fdx);
    lstr = load(fileName);

    nrLabeledFrequencies = [];   
    lteLabeledFrequencies = [];
    signalLength = max(size(lstr.resampledData));
    signalDescriptors = lstr.labeledBandwidth.SignalDescriptors;
    for sdi = 1:numel(signalDescriptors)
        signalCf = signalDescriptors(sdi).CenterFrequency;
        carrierCf = lstr.labeledBandwidth.CenterFrequency;
        % Create default label with the minimum bandwidth for the standard
        if isa(signalDescriptors(sdi),"hNRPSSDescriptor")
            bw = 5e6;
            type = "NR";
            labeledFrequencies = nrLabeledFrequencies;
            nrLabeledFrequencies = [nrLabeledFrequencies,cf];
        else % isa hLTEPSSDescriptor
            bw = 1.4e6;
            type = "LTE";
            labeledFrequencies = lteLabeledFrequencies;
            lteLabeledFrequencies = [lteLabeledFrequencies,cf];
        end

        % Do not create labels for the same signal type within the minimum
        % bandwidth of an existing label.
        if ~isempty(labeledFrequencies) && min(abs(cf-labeledFrequencies)) < bw/2
            continue
        end

        labels(1) = signalCf - bw/2 - carrierCf;
        labels(2) = signalCf + bw/2 - carrierCf;
        lss.setLabelValue(fdx,"SpectrumSensing",[0,signalLength/lstr.sampleRate],labels,type);
    end
end

Label 5G NR Carrier Waveforms

Open the Signal Labeler app.

signalLabeler

To import the signals and labels into the app, select Import from the Labeler tab in the toolstrip. From the dropdown menu, select From Workspace. In the dialog that appears, select the labeledSignalSet object lss from the workspace browser and click on Import and Close.

Select a member from the Labeled Signal Set Members table in the app and click on the Plot checkbox to plot the signal in the app's axis. In the Display tab, select Spectrogram from the Views options and deselect the Label Viewer. This enables the spectrogram axis to be visible and removes the label viewer axis.

In the Labeled Signal Set Members table, click the arrow to the left of the name in the table to expand the children. The children include the member name and corresponding label definitions. Right-click the label definition Spectrum Sensing and select Plot All. Then, expand the labels associated with that member.

A label, represented by a bounded and colored rectangle on the axis, is added for each NR and LTE signal detected during the scan for wireless signals step. Each rectangle is centered on the center frequency where it was detected and has a bandwidth equal to the minimum expected bandwidth for the standard. In some cases, depending on the tuning of the preamble detector, the same LTE signal might be detected with small frequency offsets.

Due to the flexibility of the 5G NR standard, any detected 5G NR PSS indicates the presence of an SSB, but does not indicate the center of the carrier waveform. To precisely manipulate the label location and bandwidth, first select the vertical zoom control from the axis, then zoom in on the spectrogram and NR label.

Enable editing labels by clicking on the mouse icon on the axis toolbar.

Select the generated label by clicking in the center of the label. Move the generated label to one edge of the carrier waveform by clicking the label and moving the cursor to the edge of the waveform. If the detected NR SSB has no associated carrier waveform, the signal is obscured by noise, or the detection is a false positive, right click on a label in the Labeled Signal Set Members table to delete it.

Then click on an edge of the label and move the cursor to include the full carrier waveform.

Double-click the label, or press Enter to commit the label instance.

Label LTE Carrier Waveforms

Any detected LTE PSS are in the center of an LTE carrier waveform. Click and move each edge so that the label includes the carrier waveform.

Label Unknown Waveforms

Finally, label any visible signals that are not NR or LTE as Unknown. Enable adding labels in the axis by clicking on the draw labels icon in the axis toolbar. Click and drag across the limits of the unknown signal.

Select Unknown in the label value dropdown box. Commit the label by double-clicking or pressing the Enter key.

Apply Labels and Save

When you have finished labeling a member, deselect the Plot check box and toggle on the next member in the Labeled Signal Set Members table. Repeat for all captured bandwidths.

When you have finished labeling signals, export the labeledSignalSet object back into the workspace. In the Labeler tab, click the Export button and select To Workspace. Name the labeled signal set variable.

Compute Spectrogram

Once you have all the labels of interest for your set of signals, it is useful to preview how they look outside of the Signal Labeler app. You can create a timeFrequencyMask object to show the spectrogram of the signal as well as the signal locations. This operation is useful for showing what might be fed to a machine learning model.

% Gather necessary data from the labeledSignalSet object
[x, info] = lss.getSignal(1);
fs = info.SampleRate;
labelTable = lss.Labels.SpectrumSensing{1}

labelTable=2×3 table
    0    0.0500    2.1393e+07    3.8640e+07     NR
    0    0.0500    5.6219e+07    8.8060e+07    LTE

% Create a timeFrequencyMask object and plot the spectrogram and mask
mask = timeFrequencyMask(labelTable,tfOpts,SampleRate=fs, ...
    OverlapAction="prioritizeByList");

plotMask(mask,x(1:1e6))

Further Exploration

You can decode the bandwidth of an LTE carrier waveform using LTE Toolbox™. For more information, see the Cell Search, MIB and SIB1 Recovery (LTE Toolbox) example. Consider using LTE Toolbox to automatically set the bandwidth of the label for each detected LTE signal.

For NR, you cannot decode the bandwidth of the carrier waveform using 5G Toolbox™. However, decoding the MIB can confirm that a carrier waveform exists and can give the location of the SIB1 within the bandwidth of the carrier waveform. For more information on decoding the MIB, see the NR Cell Search and MIB and SIB1 Recovery (5G Toolbox) example. Consider using 5G Toolbox to automatically place the label of each detected NR signal within the bandwidth of the carrier waveform.

For more information on using the labeled data to create a network to detect 5G and LTE signals, see the Spectrum Sensing with Deep Learning to Identify 5G, LTE, and WLAN Signals example to train a semantic segmentation network to identify 5G NR and LTE signals in a wideband spectrogram.