Project GFS: The GFSK Receiver

This project is a compulsory part of the examination for the Implementation of Digital Signal Processing course at the University of Twente. The goals of this project are:

Preparation

Read the document Some Background on GFSK Modulation carefully. It contains the theory and some practical remarks that you will need in the rest of this project. Consult also the lecture slides on GFSK modulation as presented as part of the CORDIC theory.

The GFSK system considered looks as follows:

As compared to the setup presented in the CORDIC lecture, the system has two transmitters. The first is transmitting at the wanted carrier (IF) frequency 1 MHz. The other is an interferer signal, for example originating from a device operating in another radio channel; it transmits at a frequency of 3 MHz. The wanted signal has an amplitude of 1. The interferer has an amplitude amp_interf that can be controlled by the testbench user. This amplitude is zero by default, which amounts to the interferer not being present. Interferers are quite common in radio communication. Many standards have multiple channels in some frequency band. The systems should be designed in such a way that communication in adjacent channels should not interfere too much.

The system setup acts as testbench for the GFSK Demodulator, the design of which is the topic of this project. As explained in the background document, the demodulator is built from a chain of 4 signal-processing blocks, as follows:

Files and Directories

Go to your home directory and fetch the files for this project with: 
get-module idsp-gfs idsp-gfs
Note: The first argument of get-module is the project name, the second the name of subdirectory in your file system. So, by issuing the command several times with a different second argument, you will be able to make multiple copies of the distribution.

Three subdirectories of idsp-gfs will be created:

A multitude of files are involved in this project. They will be presented gradually, at the moment that they will be needed.

Change to subdirectory arx and run make. This ensures that the database is consistent by generating the C++ and VHDL versions of the Arx source files present in the subdirectory. At this moment, you don't need to understand contents of the source files.

Now change to directory cpp and run make in that directory as well. You can ignore warnings originating from the IT++ source files.

The cpp directory has IT++ code for both the modulation and demodulation as well as the C++ code corresponding to the four demodulator blocks. After running make 5 testbenches (5 compiled executables) are created from a single generic source file tb_ber.cpp:

Each executable will perform a bit-error rate (BER) simulation as follows. The noise level is given by the signal-to-noise ratio (SNR). The wanted signal has an amplitude of 1 and will not change throughout the entire project (as opposed to a real-life situation where the received signal can vary many orders of magnitude and gain control in the analog front end keeps the signal in a relatively small range). The SNR is computed with respect to that fixed wanted signal level. So, SNR = 0dB amounts to having as much signal power as noise power in the frequency band of interest. In a loop the SNR is gradually increased (the noise power is decreased) in steps of 1 dB until the BER drops below 1.0e-4 (one error in 10.000 bits). At a specific SNR level, the iteration is aborted either when a maximum number of errors (1000) is reached or a maximum number of bits have been simulated (100000). If an SNR level is found with BER less than 1.0e-4, the sensitivity level of BER = 1e-3 is estimated by interpolation.

Note: In this project description, the words bit and symbol are used as equivalent notions as GFSK transmits one bit per symbol.

As a sanity check, run all executables by typing the executable name after the shell command prompt. Each simulation will display an SNR level, the BER at that level, and between parentheses the number of bits used for the computation of the BER. If everything went well, all simulations will report a sensitivity level of about 10.8 dB. Throughout the entire project this value will be a reference for the quality of the design. Any design you make should keep the sensitivity within 0.5 dB of this value, so below 11.3 dB.

The testbench executables all support the same command-line options:

Each of the 5 testbenches writes files in an output directory of its own, named output_slicer, output_mixer, etc. The following files are written:

Exercise GFS-1: Familiarization with GFSK

Study the background-information document in detail and try to match it to the IT++ models of both transmitter and receiver as given in files gfsk.h, gfsk.cpp, tb_ber.h, and tb_ber.cpp. The simulation operates bit by bit. For each bit, the modulation function is called. Due to Gaussian filtering which spans multiple bits (how many?), the modulator needs to keep track of its history. Also at the demodulator side, the simulation advances in steps of one bit and history matters e.g. for the FIR filters.

Now run the following simulation (SNR = 100 dB basically means that there is no noise):

tb_ber_itpp -d -b 1000 -s 100

Then, launch Matlab (command matlab) and in Matlab, plot the frequency spectrum of the modulator output with:

load output_itpp/mod.out

periodogram(mod,[],2^13,8e6)

Inspect the spectra of intermediate signals working your way step by step in the signal processing chain. Look at the ADC output, mixer outputs and LPF outputs. Explain what you observe referring to the theory.

Inspect the time signals dam.out and slc.out as well as the input bit stream. Do you recognize the input bit stream in the slicer output? How much latency does the system have (how much "garbage" comes out of the slicer before the first input bit is output? How large is the amplitude of dam.out and how does this relate to the theory presented in the background document?

Repeat the simulation with an interferer amplitude of 1.0. Comment using the spectra as well as the time-domain plots.

Exercise GFS-2: Bit-Error Rate Simulations

The correct synchronization is essential for the BER performance. Two parameters that matter are the system latency, pipeline depth (for the bits stored on behalf of BER calculations) or symbol offset and the slicer offset. The optimal values for these have been hard-coded in file tb_ber.h but can be overridden with the -p and -o command-line options. For testbenches tb_ber_itpp and tb_ber_slicer, investigate neighboring values to the optimal settings. In some cases, the BER will be degraded, in others the BER will collapse completely and no sensitivity level will be found. Give a few examples of both cases documenting the sensitivity numbers for each of the selected settings.

The lesson learned is that synchronization is crucial for performance. So, if you modify the design later on in a way that the latency is affected, you will need to find the optimal offset values again (inspecting the DAM and slicer outputs in the time domain may give a good clue).

Exercise GFS-3: Arx Implementation of Demodulator

Inspect the Arx implementations of all four blocks in the demodulator and compare them to their models in IT++. Describe the main differences in a few sentences per block. Perform the followng two simulations:

tb_ber_itpp -b 100 -s 100 -d

tb_ber_slicer -b 100 -s 100 -d

Plot the dam.out signals of both simulations in one graph. You should see two main differences. Mention and explain them.

For each Arx block estimate the number of flipflops used from the source code. Report and motivate the numbers.

Exercise GFS-4: Demodulator in Arx: VHDL Synthesis

The distribution of this project comes with support for VHDL. An entity gfsk.vhd is provided that instantiates all blocks coming from Arx as well as a clock generator for the different clock rates necessary. Word lengths are centrally administered in a file pk_gfsk.vhd. The clock-generator is a "quick-and-dirty" solution. In a real design, clock division would be implemented in a more robust way (not relevant for this project).

A testbench is provided for a standalone VHDL simulation. You can run a Questasim simulation if you are curious. As opposed to what you are used, no configuration is provided as there is just one architecture for each entity. The compulsory part of the VHDL work concerns synthesis. Synthesize the design using command srun generate-design and inspect the log file when synthesis is ready. The log file contains the sizes of each of the four blocks in the demodulator. Create a table that shows for each block:

Is the number of flipflops a good indicator for the area of a block?

Exercise GFS-5: Fixed-Point Optimizations

Execute the following simulations:

tb_ber_itpp -i 7

tb_ber_itpp -i 10

You will observe that the BER performance behaves well at interferer amplitude of 7 but not at 10. Now, edit file tb_ber.h and increase the values for MIXER_WORD_LENGTH and MIXER_IWORD_LENGTH by one. The values concerned are already in the file and just need to be commented out. Recompile the testbenches by running make and repeat the above simulations. If everything went well, you should observe that the system can deal with interferer amplitude 10.

This is, however, not true for the testbenches that incorporate Arx designs. Obviously, robustness against an interferer was not taken into account when choosing the fixed-point formats in the Arx code. A redesign action is needed. In Arx, the fixed-point word lengths are provided in Arx code. Go to the Arx directory and modify their fixed-point parameters, making your way block by block from mixer to slicer and using the appropriate testbench in the cpp directory. Optimize the fixed-point parameters of each block and try to arrive in this way (close) to the smallest design which can deal with interferer amplitude 10. Each time you modify Arx code, you will need to run make in the arx and then the cpp directory. The goal is, of course, not to make word lengths larger than necessary.

Report the found values for the fixed-point parameters supported by simulation results.

Hints:

When you have optimized all fixed-point parameters, go the vhdl directory and synthesize the design. As you have changed the word lengths, you will need to modify the values in file pk_gfsk.vhd to reflect the changes before synthesis. Check your log file for errors and warnings. If you are sufficiently convinced that the synthesis was successful, make a new table with the area and flipflop count per block, as you did for GFS-4.

Exercise GFS-6: Polyphase Implementation of the FIR Filters

The demodulator contains downsampling (by a factor 4) directly after low-pass filtering. This means that a polyphase implementation, exchanging the order of filtering and downsampling is possible (see also the lecture slides).

Consider the advantages and disadvantages of a polyphase implementation in this concrete case. You can involve the following elements in your reasoning: one-to-one implementation vs. scheduled solution, filter symmetry, multiplierless design, power-area-time trade off. Do not write any code at this time. What is your final recommendation regarding the polyphase implementation? Should it be used or not?

Exercise GFS-7: Scheduled Solution

The reference design is a one-to-one implementation where the sample frequency of 8 MHz is equal to the system clock frequency. Suppose that the technology available to you allows a system clock of 80 MHz and you can do arithmetic operations (add, multiply, etc.) in one clock cycle.

Sketch on paper a data path, a bit in the style of Project MAP, that:

Assume that the controller is a sequencer that, for each clock cycle, reads a pattern of control signals from an external memory. This means that the data path can be "programmed" by changing the contents of this external memory. You do not need to be concerned about the sequencer and the memory and can assume that the control signals of your data path will be correctly supplied.

First estimate the complexity of this design: how many additions, shifts, multiplications, etc. would a data-flow graph of the entire receiver have? How would those numbers translate in a minimum of functional units (adders, multipliers, etc., FUs)? Involve the fixed-point formats to determine the word lengths that the arithmetic blocks should have.

Try to design the data path in a generic way. Avoid a "random" interconnection of multiplexers and registers. Make use of multi-ported register files instead from which FUs can read their inputs and to which FUs can write their outputs. An extreme case of such a multi-ported register file would allow all FUs to read and write their data simultaneously from the same register file.

You do not need to provide all details. A rough sketch in combination with a motivation of your design choices and design dimensioning is sufficient. Consider the choice between a generic data path for all stages of the receiver and separate data paths for each stage (mixer, filter, etc.).

Exercise GFS-8: Free Design Assignment

Make, depending on the time left, minor or major modifications to the Arx code for the GFSK receiver. Follow any of the suggestions below, a subset or a combination or just do something completely different.

As mentioned earlier, the constraint on your design should be to keep the BER performance degradation within 0.5 dB with respect to the original design. Be aware that changing the number of registers in the signal path may affect the system latency. If this is the case in your design, you will need to find the optimal synchronization parameters.

You may need to modify the C++ testbench files. Try to keep modifications to a minimum and make sure that Arx blocks are directly connected to each other, so without performing intermediate computations in the testbench.

Once your design is ready, run synthesis in order to have an impression of the area and timing. You may synthesize the entire receiver, but you may also consider synthesizing each component separately.

For debugging Arx using generated C++, consult the debug "points of attention" mentioned in the description of Project MAP. As a last resort, you could try to debug the generated VHDL.

Deliverables

Write a short report always motivating your choices and explaining the way you have reached your answers. Particular points of attention:

Grading

Go (back) to  Sabih's Home Page.
Last update on: Thu Mar 18 22:17:13 CET 2021 by Sabih Gerez.