HW/SW Co-Design QPSK Transmit and Receive Using Analog Devices AD9361/AD9364
This example shows how to implement algorithms on the Zynq radio platform that are partitioned across the ARM and the FPGA fabric. QPSK transmit and receive functions are implemented in hardware and software and mapped to the radio platform as shown in the diagram below.
HDL Coder Support Package for Xilinx Zynq
Embedded Coder Support Package for Xilinx Zynq
Communications Toolbox Support Package for Xilinx Zynq-Based Radio (this package)
In this example, the transmit and receive FPGA implementations of a QPSK system are combined into one HDL IP core and implemented on the Zynq programmable logic (PL). The data encoding and decoding is run on the Zynq ARM processor through code generation. Some control parameters are added to the FPGA IP core to show how the design can be adjusted in real time using AXI4-Lite registers accessed from Simulink.
If you have not already done so, ensure you follow the Installation for Hardware-Software Co-Design in the documentation.
Hardware Generation Model
The hardware generation model is used to develop the functionality that you wish to implement on the FPGA fabric. HDL-optimized QPSK transmitters and receivers are modelled and then implemented on the FPGA fabric of the Zynq radio platform. In this example, we transmit and receive on a single board, however, the example could be modified to work in frequency division duplex (FDD) by moving the transmit and receive center frequencies apart and using the same model on two separate boards.
Hardware/software partitioning: In general, the programmable logic of the FPGA is used for high rate signal processing while the ARM is used for slower rate, control functionality. In this example, the QPSK transmit and receive physical layer front ends are implemented on the programmable logic, as these include high rate operations such as gain control, filtering and frequency compensation. The data encoding and decoding are much slower rate and are implemented on the ARM, which decodes the message and sends to the host for printing.
The IP Core Generation Workflow used to implement the FPGA IP core and generate the software interface model has some specific requirements:
Complex inputs and outputs are not supported at the ports of the HDL subsystem. Therefore, real and imaginary signals must be modelled at the subsystem boundaries.
The data inputs and outputs to the subsystem are modelled using separate data and valid signals. The input and output clock rates of the subsystem must be equal, in Simulink the data and valid lines must be driven at the same sample rate. The valid signal must also be modelled at the input and output of the user logic.
AXI4-lite Control Port txSrcSelect: A control input
txSrcSelect is added to allow some control over the transmitted data. The
txSrcSelect port on the HDL subsystem is used to select between two different data sources for the transmitter. If the
txSrcSelect port is true, the data source for the transmitter will be a look-up table stored on the FPGA fabric and the received data should resemble the "Hello World 0XX" strings seen in the other QPSK examples. If the
txSrcSelect port is false, the data source for the received data will be the ARM processor, which will generate samples in real-time and send them to the transmitter on the FPGA fabric. The message in this case will be "*Zynq HW/SW Co-design*". The message text is taken from a workspace variable
txtStr which can be modified at compile time to change the message. Note that the length of this string must be less than 24 characters.
AXI4-lite Control Port rxStreamEnable: A control input
rxStreamEnable is added that ensures the HDL algorithm only processes data when this input is true. The
rxStreamEnable port is always true for simulating the hardware generation model. When the full design is implemented across both the FPGA fabric and ARM processor of the Zynq Radio platform, this value will be set to true only when the transmitter HDL algorithm has begun transmitting valid data, ensuring the receiver locks correctly.
You can run this model and confirm its operation. By double clicking the switch Internal Tx Switch you can select which source to transmit from. Note that Goto and From blocks are used to model the antenna connection and pass the transmitted data at the output of the transmit user logic to the input of the receive user logic.
IP Core Generation Workflow
Once you are satisfied with the simulation behaviour of the hardware subsystem, you can start the process of generating the HDL IP Core, integrating it with the SDR reference design and generating software to run on the ARM.
In preparation for targeting, you must set up the Xilinx tool chain by invoking
hdlsetuptoolpath. For example:
>> hdlsetuptoolpath('ToolName','Xilinx Vivado','ToolPath','C:\Xilinx\Vivado\2019.1\bin\vivado.bat');
Start the targeting workflow by right clicking the
HDL_QPSK subsystem and selecting
HDL Code / HDL Workflow Advisor.
In Step 1.1, select
IP Core Generationworkflow and the appropriate Zynq radio platform from the choices: ADI RF SOM, ZC706 and FMCOMMS2/3/4, ZedBoard and FMCOMMS2/3/4, ZCU102 and FMCOMMS2/3/4, ZC706 and FMCOMMS5.
In Step 1.2, select
Receive and transmit pathreference design. You can leave the reference design parameters as the defaults.
In Step 1.3, the interface table can then be used to map the user logic signals to the interface signals available in the reference design. In this example, we are only using a single channel, so the channel 1 connections should be connected to the relevant ports as shown below.
In Step 1.4, ensure the DUT Synthesis Frequency is set to a reasonable number given the baseband sampling rate of the system. In the shipping example, the sample rate is just above 520ksps, so a synthesis frequency of 1MHz is sufficient.
Step 2 prepares the model for HDL code generation by doing some design checks.
Step 3 performs the actual HDL code generation for the IP core.
Step 4 integrates the newly generated IP core into the larger Zynq SDR reference design, generates the bitstream and helps you load it onto the board.
Execute each step in sequence to experience the full workflow, or, if you are already familiar with preparation and HDL code generation phases, right click Step 4.1 in the table of contents on the left hand side and select
Run to selected task. You should not have to modify any of the default settings in Steps 2 or 3.
Software generation model and block library
In Step 4.2, the workflow generates a Zynq software generation interface model and a block library. Click the
Run this taskbutton with the default settings.
Software Interface Library
The library contains the AXI Interface block which has been generated from the HDL_QPSK subsystem. Note that this exposes only the AXI4-lite control ports, and not the data ports. The data ports are present on the Receiver/Transmitter blocks which represent the data interface between the FPGA user logic and the ARM. If you use the library blocks in your downstream models, any updates you make to your HDL subsystem will automatically be propagated to this library and then to your software generation models when you run through the workflow. In this example, the hardware generation model did not contain any SDR transmit or receive blocks so the parameters on these blocks could not be populated. When using the library blocks you must ensure to configure the parameters correctly for your application.
Software Interface Model
The software interface model can be used as a starting point for full SW targeting to the Zynq: External mode simulation, Processor-in-the-loop and full deployment. Note that this generated model will be overwritten each time Step 4.2 is run, so it is advisable to save this model under a unique name and develop your software algorithm in there. A software interface model has been provided which shows how you may decide to structure this model, see section Running the Software and Hardware on the Zynq board.
Bitstream generation and loading
The rest of the workflow is used to generate a bitstream for the FPGA fabric and download it to the board.
In Step 4.3, the workflow advisor generates a bitstream for the FPGA fabric. You can choose to execute this step in an external shell by ticking the selection
Run build process externally. This selection allows you to continue using MATLAB while the FPGA image is being built. The step will complete in a couple of minutes after some basic project checks have been completed, and the step will be marked with a green checkmark. However, you must wait until the external shell shows a successful bitstream build before moving on to the next step.
Step 4.4 downloads the bitstream onto the device. Before continuing with this step, call the
zynqfunction with the following syntax to make sure that MATLAB is set up with the correct physical IP address of the radio hardware.
>> devzynq = zynq('linux','192.168.3.2','root','root','/tmp');
By default, the physical IP address of the radio hardware is 192.168.3.2. If you alter the radio hardware IP address during the hardware setup process, you must supply that address instead.
Alternatively, if you want to load the bitstream outside Workflow Advisor, create an SDR radio object and call the
downloadImage function on the object.
If in step 1.1 you selected either the ADI RF SOM, the ZC706 and FMCOMMS2/3/4, the ZedBoard and FMCOMMS2/3/4, or the ZCU102 and FMCOMMS2/3/4 radio device, create an AD936x radio object.
>> radio = sdrdev('AD936x');
If in step1.1 you selected the ZC706 and FMCOMMS5 radio device, create an FMCOMMS5 radio object.
>> radio = sdrdev('FMCOMMS5');
Download the bitstream using radio object interfacing the selected radio device.
>> downloadImage(radio,'FPGAImage', ... 'hdl_prj\vivado_ip_prj\vivado_prj.runs\impl_1\system_top.bit') % Path to the generated bitstream
Constructing the QPSK software interface model
A software interface model has been provided which shows how you could modify the generated model to set it up for the QPSK example. This interface model allows you to run the model in
Monitor & Tune mode or fully deployed.
Note that in the software interface model, the AD936x Transmitter block handles the frame to sample serialization.
Setting up the software model to run on the ARM processor
The application model has been set up following the guidelines in the Hardware-Software Co-Design Workflow, section Configure Software Interface Model.
The model has been configured to run using the Xilinx Zynq-7000 Based Board target for the following hardware: ADI RF SOM ZedBoard and FMCOMMS2/3/4 ZC706 and FMCOMMS2/3/4/5
If you are using the ZCU102 and FMCOMMS2/3/4, double click the Selected Hardware Board Target block to change the configuration to use the Xilinx Zynq UltraScale+ MPSoC ZCU102 IIO Radio board target.
The model is continuously transmitting and receiving data, so it has been configured to run from the Transmit interrupt. This ensures that the ARM and the FPGA are running in synchronisation and means that the software will be driven by a schedule tick at the frame rate.
To ensure efficient use of the ARM processor, the transmit frame size has been increased to 10000 by concatenating 50 frames at a time using a For Iterator Subsystem and MATLAB Function block. The subsequent Rate Transition block forces the data rate from the For Iterator Subsystem to match the transmission data rate.
The transmitter underflow has not been connected to a Stop block as underflows may happen when the source switch is toggled when running in
Monitor & Tunemode. This is as a result of the increased processing required by
Monitor & Tunemode register writes. See the section below on running the model using UDP blocks to control the hardware.
rxStreamEnableAXI4-lite control port is delayed by 2 frame periods. This ensures the transmitter is running before the receiver for the most robust performance.
As the QPSK receiver contains a downsample by two operation, the valid signal at the output of the user logic is used to reduce the effective sample rate to half that of the clock rate. In the model, the frame rate of the ARM processor is therefore set to half the sampling rate multiplied by the frame size. The frame size for the receiver is half that of the transmitter for this same reason.
The Receive Timeout has been set to 200 s, which is small in relation to the frame period.
Running the Software and Hardware on the Zynq board
Monitor & Tune mode allows you to control the configuration from the Simulink model. Once the design is running, switch between sourcing data from the ARM or the FPGA fabric by toggling the Tx Switch.
Once the ARM has decoded the QPSK message, it sends the result back to the host over the Ethernet link using the UDP send block found in the software interface model. The UDP send block has been configured using the default IP address of the host '192.168.3.1'. If you altered the IP addresses during the hardware setup process, you must supply that address instead. A simple UDP receive model has been supplied which can be used to receive the decoded data and display the result in the diagnostic viewer.
You can also fully deploy the design to run on the board, disconnected from Simulink. In the Simulink toolbar, click Build, Deploy & Start. In this mode you will not be able to tune parameters.
Using UDP blocks to control the user logic
Monitor & Tune mode requires some overhead to be included in the software running on the hardware to deal with communication between the host and the board. As seen in the software interface model, switching between the ARM and FPGA transmitter sources resulted in transmit underflows. An alternative interface model has been supplied which shows how UDP blocks can be used as an alternative switching mechanism which requires less overhead.
In this model, the switch has been replaced by a UDP receive block which will be able to receive UDP packets and output the source choice value. Some further modifications have been made to the model.
The transmitter underflow has now been connected to a Stop block which will cause the model to exit whenever an underflow is detected. A Step source has been used to gate the overflow signal for the first ten frame periods as the hardware starts up. Some underflows may be experienced as the transmitter performs buffering, and then again when the receiver starts two frames later. The receiver initialisation will result in a one-time load on the processor that may cause underflows.
The UDP Receive block has a default output value of 0, so an inverter has been placed between the UDP Receive block and the multiplexors to ensure that the transmitter starts up with the FPGA transmit source. In the Simulink toolbar, click Build, Deploy & Start. Once the model has deployed to the hardware, a UDP transmitter source can be used to drive the transmit source selection.
A simple UDP transmitter model has been supplied which can be used to drive the transmit source select on the hardware over the Ethernet link. The UDP send block found in the transmitter model has been configured using the default IP address of the radio '192.168.3.2'. If you altered the IP addresses during the hardware setup process, you must supply that address instead. A UDP receive block is included in this model. It can be used to receive the decoded data and display the result in the diagnostic viewer.
This model has been configured to run for a single step and send a single UDP packet containing the source select value. Set the source select to your desired source and click play to send a UDP control packet.