Transmission of an image with the USRP (matlab/windows)

1. Objective of the demonstration

The aim of this demonstration is to transmit an image from a PC connected to an USRP in transceiver mode to another PC connected to a second USRP. The objectives are the following ones:

1. Convert the image into a sequence of QPSK/16-QAM symbols
2. Generate a linear modulation of these symbols at a rate 200 kHz
3. Transmission and reception of this signal
4. Estimation of the transmitted symbols by the receiver
5. Reconstruction of the transmitted image from the estimated symbols

Two constellations are tried in this demonstration: QPSK and 16-QAM. The baudrate of the transmission is reduced by a factor 2 using the 16-QAM symbols compared to the QPSK symbols. But as illustrated, this baudrate gain induces an increase of the bit error rate of the transmitted symbols.

Only the steps 2 to 4 are described in this page. A video at the end of this page shows the whole transmission.

2. Transmitted signal generation

2.1. Symbols mapping

To allow correction of the channel impulse response variation and RF imperfections, training sequences are included in some parts of the transmitted symbols sequence, as illustrated on the image on the right.

The information symbols are split in frame of 1200 symbols. At the beginning of each frame, a training sequence of 150 symbols is inserted to estimate the channel impulse response and to synchronize. This training sequence is known from the receiver.

A training sequence of 300 BPSK (real) symbols is also inserted before the first frame. The objective is to estimate and to compensate the frequency offset of the received signal. Note that the receiver does not need to know the value of these symbols to perform this compensation step.

No error protection code has been used in this demonstration.

2.2. Signal generation

The obtained sequence of symbols is then linearly modulated with a square-root cosine shaping filter at a symbol of 200 kHz and with an excess bandwidth factor of 0.2.

An over-sampled version at 1MHz of this signal is passed through the USB port to the transmission module that interpolate the signal, modulate it and transmit it.

3. Signal transmission and reception

3.1. Signal transmission

The signal is modulated in the WiFi channel #3, around the frequency 2422MHz. The transmission USRP transmit this signal several times, each replica being separated from the previous one of 2 times the signal of interest length, so that the transmission least around 20 seconds.

3.2. Signal reception

During the 20 seconds the transmission lasts, the receiving module connected to the second PC has to get this signal. An acquisition process to listen to the Wifi channel #3 is then launch during 4 times the time of the useful signal.

4. Estimation of the transmitted symbols by the receiver

4.1. Burst selection

The received signal on the PC connected to the receiving module is hence a sequence of burst, each being a replica of the signal of interest. The first treatment is hence to select one of this burst, to then process it and to extract the information symbols. This step is manually done.

4.2. Frequency offset estimation

The second treatment consists in estimating the frequency offset of the received signal. The 300 BPSK training sequence is therefore exploited. The Fourier transform of the square of this part of the received signal has 3 peaks. These peaks are shifted and not centered on 0. The fundamental is at a frequency equal to 2 times the frequency offset. Once estimated, the frequency offset is compensated.

4.3. Adaptive filter

The obtained signal is then filtered by its adaptive filter (the shaping filter use by the transmitter). This step leads to a maximization of the signal to noise ratio of the signal.

4.4. Synchronization and equalization

The decoding of the transmitted frames can now start. For each frame, the first step consists in synchronizing temporally. In this demonstration, the received signal is oversampled by a factor 5, meaning that 5 sequences of received symbols are available.
For each sequence, the channel impulse response is estimated thanks to the training sequence inserted at the beginning of the frame, as the estimation error. The synchronization is done by selection the sequence for which the estimation error is the smallest one.

During the synchronization step, the channel impulse response is also estimated. As illustrated on the left, the channel can be considered as Gaussian. The channel equalization is hence simply reduced as the division of the received signal by the impulse response higher coefficient. The symbols estimation is then performed with an optimal decision process.

5. Illustration

Comments are in French, subtitles in English are available: