<aside> 📡 The scope of this project was to develop a software implementation of a radio receiver that was capable of processing FM broadcast signals in real-time. Our objective was to create a system that could accurately extract IQ samples and demodulate FM signals, extract mono and stereo audio channels, and decode Radio Data System (RDS) information.
</aside>
Using radio-frequency hardware such as the RF dongle, and Raspberry Pi 4 in parallel with initial Python model to C++ to extract radio information from a chosen FM channel. We used these signals to successfully demonstrate our system's functionality in a physical setting with real time data.
This model was used to process mono and stereo audio data, and RDS.
Mono
The mono path takes in the FM demodulated signal with the specified IF sample rate depending on the mode of operation and extracts the mono audio channel which spans from 0 to 15KHz, carrying the sum between the left and right audio channels. The signal flow diagram consists of resampling, low-passing, and convolution.
Stereo
The stereo path includes carrier recovery, channel extraction, and stereo processing. Carrier recovery involves extracting the 19kHz pilot tone and passing this signal to the PLL and NCO. This ensures that the frequency of the local oscillator accurately matches the carrier, which is crucial for the purity of the final signal. This way, the original audio signal can be recovered when mixed with the message signal. The message signal is the output from the 22kHz to 54kHz band-pass filter in the channel extraction stage. Finally, stereo processing merges the separate left and right channels with the final output of the mono path, reproducing the original audio.
RDS
The RDS path extracts information about the broadcasted information, such as the program identification code, program type, and program service name. The overall flow of RDS consists of first extracting the RBDS signal in the FM channel using a 54KHz to 80KHz bandpass filter, and then recovering the 114KHz carrier using a bandpass filter along with the phase lock loop (PLL). Then, the final signal to be processed is obtained after demodulation, where the carrier is mixed with the message signal, filtered, resampled, amplified, and recovered. The final output is obtained by a frame synchronized to match the block information to extract the message from each block's information bits.
Figure 1. Error due to unsuppressed Q samples.
Figure 2. Proper output with Q (pink) suppressed and I (purple) with higher energy.
Figure 3. Output after the Phase Lock Loop (PLL)
To extract data accurately, the impulse responses needed to be sampled at the peaks of the waves. After re-determining the offsets every few blocks of data, we were able to get a more accurate reading and the IQ constellation plot split as expected.