Lab 1

Overview

Last week you used an integrated program, gqrx or SDR#, to control your SDR and listen to commercial FM, along with a range of other frequencies. 

This week we'll get a little closer to the hardware, and learn how to control the SDR's more directly. We'll use MATLAB's support for the rtl-sdr to read data directly, so that we can examine and decode it. We'll capture AM signals in the airband.

Aims of the Lab

This week we'll look at other ways to control your SDR to capture AM airband data with MATLAB and process it. You will capture a signal, display it's spectrum as an image, decode the signal, plot it, and play back the audio.

Install Matlab

PLease install Matlab 2025. If you have earlier versions, please upgrade. Matlab 2025 is needed for the support of the RTL-SDR .com V4 dongle.  Make sure you have the following toolboxes installed:

• Signal Processing Toolbox

•  Communication Toolbox

•  DSP systems Toolbox

•  Instrumentation Control Toolbox

•  Optimization Toolbox

•  Statistics and Machine Learning Toolbox

If you already have Matlab 2025, you can add the toolboxes within MATLAB by clicking on the “Add-ons” button. This brings you to a pulldown menu to manage and explore add-ons.

MATLAB Hardware Support for the RTL-SDR

Choose again Add-Ons Menu, click the “Get Hardware Package Support” entry. This brings up a page for all of the different types of hardware MATLAB knows how to work with, including several sdr's, including the rtl-sdr that we are using. Scrolling doen the pgae, you should see somethign like this:

Click the install button, to install the package. If you already have it installed, it may say “Manage” because . Follow the instructions on that page, including plugging your sdr in, and running the initialization process from the pop up window. At this point you should be all set. If you are a windows user, you already have the driver installed, so you can skip that step.

which is avaliable here:

Read_RTLSDR.m

The rtl-sdr device  is freed after it is used, so that you can run gqrx. However, you need to quit these before you run the m-file again, or MATLAB hangs, since it can't get access to the device.

Collecting Data, and Computing Spectrum and Spectrograms

Let's collect data around the FM band, where we know what spectrum to expect

capture 1 second of data with using: Read_RTLSDR(94.1,2.4e6,1e5,24,25);  The center frequency is 94.1Mhz, sampling rate 2.4MS/s, each buffer from the sdr is 10000 samples, we collect 24 buffers, and set the SDR gain to 25db.  48*10000/2.4e6 = 1 seconds!

Now, compute the spectrum using the FFT. We will use fftshift to center the spectrum.

Let's display the signal. The frequency range is between 94.1-1.2 and 94.1+1.2 MHz and will have 2.4e6 samples. 

The spectrum should look like this:

The spectrum will look very noisy, and will have high resolution. The resolution is inversly proportional to the window size. So for 1 second, the resolution will be 1 Hz, which is too much! To obtain better signal-to-noise ratio in the spectrum, we will use the matlab function pwelch.

This function breaks the signal to (potentially overlapping) windows, computes their power spectrum and then averages the result. The spectral resolution will be determined by the size of the window and the signal to noise ratio will improve with more averaging.  The default windowing function used in the function is a rect window. But it is possible to use other windows, like hamming, which will result in less ringing in the spectrum. 

Use a window of 2400, which is 1ms window and a spectral resolution of 1KHz.  Make sure to add: "centered" to indicate that the spectrum is double sided, and centered at zero.  The spectrum will be much clearer!

The spectrum should look like this:

It is also useful to plot something like the waterfall plot that you see in gqrx. This is called a spectrogram. Similar to the welch method, we compute the spectrum of blocks of the signal. But instead of averaging them,  we display it as an image of how the spectrum changes over time. While Matlab has several implementations of spectrograms, we will use this basic spectrogram program that is provided here

msg.m

It is msg.m for “my spectrogram”. 

Compute and display the spectrogram of the fm941 signal you collected:

The spectrogram should look like this:

This takes an input signal starting at sample n0, and computes the spectrum of nt segments of the signal, each of length nf. The result is displayed as an image, with time going horizontally, and frequency vertically. The center frequency is in the middle of the plot. To look at the spectrum for the entire 1 s signal

I had lots of luck listening to Remote Comm Air Ground (RCAG) frmo Mt. Tam from Cory. In particular 353.500 and 323.00

We will also hear traffic from San Francisco (SFO), Hayward (HWD), as well as NORCAL Approach, which coordinates the airspace. The frequencies for SFO are

and Hayward are

and NORCAL Approach are

Choose a frequency where you might expect to get a signal. The ATIS frequencies would be agood initial candidates, because these continuously transmit information about the airport, and how to contact them.  But I was not able to capture them from my office. The other frequencies, such at the air traffic control frequencies, are more interesting, but are not always in use. Often a transmission lasts just a few seconds, and can be hard to capture.

One busy channel that has a strong signal around near Berkeley is 353.5 MHz, from Mt. Tam. Frequencies in this range are a good match for your SDR antenna. At 353.5 MHz a quarter wave antenna is 24 cm, or about 9.5 inches. Set each antenna to that length and angle at about 120 degree from each other.  

Use gqrx or to see if you can find any activity. Here is an example

You see a couple of frequencies in use. The one that is on continuously is an ATIS signal. The others are planes and towers talking to each other. You can adjust the LNA gain if necessary.

Once you know there is a signal out there, you can start capturing data.  Close gqrx or SDR# to free up the SDR, and then capture 8 seconds of data with, for example: Read_RTLSDR(310.8e6,2.4e6,1e5,192,25);  The center frequency is 310.8Mhz, sampling rate 2.4MS/s, each buffer from the sdr is 10000 samples, we collect 192 buffers, and set the SDR gain to 25db.  192*10000/2.4e6 = 8 seconds!

where I have chosen 310.8 MHz, which is Norcal Approach/Departure South. We are setting the frequency exactly to the channel we are interested in to make the processing easier. This is generally a bad idea, as we will see shortly. The data file is available at

ab310.mat

You can use this file if you are having trouble finding signals to capture. Another signal you can usually see is the Oakland ATIS, as it transmits all of the time, but is more challenging to capture.

The first thing to note is that just 8 seconds of RF is a lot of data! It is hard to tell if we have anything, or how to extract it. What we will do is to compute the spectrogram. 

Unless you have a very big display, you'll get an error message that the image doesn't fit, and was scaled down. The result looks like this

This is 8 seconds of data. The horizontal axis is time, 0 to 8 seconds, and the vertical axis is frequency, -1.2 MHz to 1.2 MHz. The signal we want is right in the middle at 0 MHz. The signal starts at about 2 seconds, and continues to about 7 seconds. Depending on your luck, you may see several signals at other different frequencies.

For now we'll just focus on the signal at 0 MHz. The audio signal has a bandwidth of about +/- 5 kHz, so we need to reduce the sampling rate from 2.4 MHz to 10 kHz. This is called decimation, in this case by a ratio of 2.4e6/10e3 = 240. A lowpass filter first limits the bandwidth to +/- 5 kHz, and then takes every 240th sample.

If we plot the magnitude of the decimated signal with

we see

This is a classic AM signal. There is an offset of about 1 unit from the DC feedthrough because we are receiving right at zero frequency. We'll look at fixing this below. Then the AM signal starts at about 1.5 seconds. What you see is the AM offset that biases the audio waveform to make it always positive. On top of that you see the actual audio signal.

Let's plot the spectrogram of the decimated signal!  Remember, the signal is decimated, so the sampling rate is 10KHz.  Choose a window size of 200. This will give you a spectral resolution of 10000/200 = 50Hz. 

You should be able to see the zero frequency DC feedthrough. You should also see the strong DC signal of the AM modulation, and the symmetric spectrum of the speech signal!

You can play the signal with

Soundac scales the waveform to be less than one, which is what sound() requires to avoid clipping, and 10e3 is the sampling rate after decimation. We haven't suppressed the AM offset, because your audio system can't reproduce it anyway.

• Based on Lab developed by John Pauly