projects

Hardware/software projects
2022 - 2024
rp2040 -- FPGA -- analog

rp2040 Microcontroller
- C-SDK -- The C-SDK is open source, has good examples and is very well documented. The C-SDK directly supports PIO assembly language, DMA transactions, and will support the highest interrupt rate of the development systems. The instructor for ece4760, Van Hunter Adams, has produced many cool C applications for the rp2040.
  My Applications:
  - ece4760 Support
    General utilities and information relevant to ece4760 at Cornell.
    All examples and student code are written using the Protohreads framework for multitasking.
    - Protothreads 1.3 (written by Adam Dunkels) to be used 2024
      A port to the rp2040, extended for two cores with support for hardware locks, core-safe semaphores, and core-to-core hardware FIFOs. Protothreads provides a simple, cooperative, thread environment with very fast context switch and low memory overhead. Modified thread dispatcher implements priority scheduling in addition the default roundrobin.
      - Protothreads 1.1.2 (written by Adam Dunkels) used 2022-23
    - Visual Studio configuration.
      Configure VS for editing, programming, debugging using the Pi Debug Probe
      - Add program/reset button
        A trivial hardware and software addition simplifies programming.
        Superceded by using the debugger.
    - WIFI setup.
      The physical considerations and software necessary for WIFI.
    - Fixed point arithmetic
      Somewhat faster the floating point on the rp2040. We implemented s15x16 (signed, with 15 bits of integer, and 16 bits of fraction) and s1x14 fixed point.
  - Graphics
    Graphics input and output using 640x480 VGA, with optional joystick or encoder wheel.
    Mouse interface is also shown in the USB section.
    The serial interface is not really graphics, but plays around with color and formatting.
  - External Memory
    Everyone needs more memory. FRAM works well for simple files with bandwidth requirments below about 1.5 Mbyte/sec.
    SD card with FatFS has a larger program size overhead, but is compatable with PC read/write of huge files.
    USB FLASH drives are exzplained the the USB MSC example below.
  - LWIP on PicoW
    LWIP is the supported TCP/IP stack for the PicoW.
    LWIP has lots of options and somewhat limited documentation.
    This section has lots of examples to exercise some of the options.
    Opotions include bringing up the PicoW as an WIFI access point or WIFI station and
    data transfer by UDP or TCP.
    - TCP protocol
      - AP/HTML server on picoW to Browser as control/data display.
        Using the rp2040 as an access point with web server allows a cellphone or
        other web browser to act as a control panel for applications running on the rp2040.
        
        (depricated) Browser as a control panel for rp2040. (version 2)
        
        (depricated) Standalone WIFI access point and web server (initial version)
        This is an example from the C SDK repository that I used to understand more of LWIP.
      - AP/HTTP server on picoW to client/viewer on picoW.
        Use HTTP protocol to serve interactive, text, pages to another picoW.
        The intent is a remote control panel and data display.
    - Network Time Protocol
      - LWIP and VGA display of NTP and RTC
        The LWIP and Hunter's VGA both use significant hardware. With care they are compatable and run at the same time.
      - Network Time Protocol (NTP) and the onboard real-time clock (RTC)
        The NTP can be used to automatically set the rp2040 RTC, perhaps daily, to maintain accuracy. The Pico board clock oscillator is rated at ±30 ppm, or accuracy of 3x10^-5. Since there are 86400 seconds/day, this means up to 2.6 sec/day drift.
    - UDP protocol
      - Symmetric send/receive between two picoWs (access point to station)
        Each picoW can send or receive packets.
        The code running on the two nodes is almost identical.
      - Audio-rate UDP from picow to picow (access point to station)
        Sending real-time data from picow to picow with no router involved.
      - Data array UDP send/receive (station to station thru hotspot)
        Sending an array of data allows performance testing as well as understanding of data sizes, and dealing with packet efficiency.
      - UDP send/receive from desktop (station to station thru hotspot)
        UDP protocol is a simple and fast, but does little error checking. It is therefore useful for data streams where a bad value is not a show stopper. You might send music but not code.
      - UDP send/receive pico-to-pico. (station to station thru hotspot)
        A simple scheme for figuring out IP addresses on the fly allows two picos to send data to each other.
  - USB on Pico
    USB uses the tinyUSB implementation and is fairly big.
    There are lots of ways to use it. Examples are your friend.
    - USB HID host with mouse and keyboard
      Connecting a mouse and keyboard with video display of results.
    - USB host with HID and MSC
      USB flash drive, Keyboard, mouse
    - USB device
  - Amusing (to me) computational projects
    These may have little general appeal, although the DMA computing machine could be a co-proessor.
    - DMA computing machine.
      The DMA subsystem is capable of running a compute-universal system. This implementation uses three channels for fetch/execute, and a list of DMA control blocks as a program.
    - Random number generation.
      Using the ROSC to make reasonable quality random numbers. Routines for integers, fixed-point uniform and normal distributions, and for single bits.
    - Lattice-Boltzmann fluid flow simulation
      A strange hybrid of finite difference and cellular automaton using fixed point arithmetic for speed. The fixed point system used is s1x25. The shorter, and faster, 16-bit fixed point did not have enough precision for stable solutions.
    - 16-bit floating point,
      with similar bit layout to the IEEE standard FP16, but without infinities or NANs. The reason for doing this annoying exercise is to see if ODE solvers run faster in limited precision floating point than in fixed point. The 16-bit floats have a dynamic range similar to s15x16 fiexed point, but are not faster then standard 32-bit floats.
    - rp2040 PIO control
      Experiments to see if modifing the code of a running PIO channel is feasable.
    - rp2040 Interpolator hardware
      - DDS and FM waveform synthesis using interpolator peripherial.
  - DSP
    More numerical flogging of the M0 architecture, with an emphisis on speed using 16-bit fixed point.
    The IIR and FIR sections use design implementations which run completely in C on the the M0, at the
    cost some slightly non-optimal designs.
    - IIR filter designer
      Using s1x14 fixed point format. Lowpass, bandpass, and highpass filters design. Plots the Bode plot of the designed filter, and allows the new filter to be used in realtime for audio input.
    - FIR filter designer
      Using s1x14 fixed point format. Lowpass, bandpass, and highpass filters design. Plots the Bode plot of the designed filter, and allows the new filter to be used in realtime for audio input. Also a version for linear frequency plots.
    - Speech compression/playback from IIR filterbank.
      Speech sampled at 12.8 Ksample/sec is sent through 32 MEL-distibuted IIR filters.
      The filter powers are sampled every 20 mSec and used to reconstruct speech using DDS.
    - FFT and FilterBank spectrograms
      FFT compared to lower resolution MEL spectrum for speech input.
    - FFT Spectrogram
      Plotting the power spectrum, log power spectrum, and spectrogram of voice signals. The spectrogram actually has enough info to decode what is being said.
    - DSP development
      IIR filters in fixed point s7x24 format. Butterworth lowpass and bandpass filter.
  - Sound Synthesis
    Since we can do DSP, can we use it to make sounds we might actually want to listen to.
    FM synthesis is very 1970's synth kind of sound.
    The Karplus-Strong algorithm actually solves the wave equation on a stirng in realtime to make
    physically reasonable stringed instrument sounds.
    - FM synthesis of sounds.
      Some FM synthesis even sounds like music. Widely used in the 1970's, '80s, and even into the 90's. for music and sound effects.
    - Karplus-Strong strings
      Plucked and bowed string physical synthesis. An approximation of the actual PDE of a string.
- MicroPython -- I played with this system for a couple of months just after the PICO was announced. Python is a very nice environment, but fairly slow for real-time programming. I got DMA running, but it was tedious because Python really does not want you to mess directly with memory. In the end, we decided to move to straight C-SDK for development.
  Applications:
  - A simple program to sample the ADC and copy the values to a PWM channel at 500 Ksamples/sec using zero cpu cycles.
  - I figured out a DDS system using the settable DMA transfer clock to produce sinewaves on a pwm channel with no cpu cycles.
- MBED/Arduino -- The MBED framework (arduino site) allows use of a simple RTOS on core0, while running bare C-SDK on core1. The tool chain is installed in the usual trivial Arduino way. Speed is good and the RTOS is quite handy. It is a bit confusing because there are three different peripherial libraries to handle: Arduino, MBED, and C-SDK. I completely eliminated the Arduino constructs and used a mixture of MBED and C-SDK. Using the PIO subsystems required an external assembler for the PIO assembly language and was a little inconvenient. In the end, we decided to move to straight C-SDK for development.
  Applications:
  - ADC and PWM DAC using DMA with
    ADC>DMA>PWM on Core0
    DDS>DMA>PWM on Core1
    The DDS algorithm is completely carried out by a DMA channel and the PWM DAC.
  - A PIO input event time/duration capture which works up to about 10 MHz.The RP2040 has no "input capture" peripherial that uses hardare to grab a time stamp for an external event (edge on i/o pin). Both the AVR and PIC32 that I have used can capture times in hardware, and I find it useful. The PIO can be used to implement a fast timer/counter, detect i/o pin edges, and log the time stamps at full bus rate to a 8-slot hardware FIFO.
  - PIO based stepper motor drivers that took high level move commands from the cpu. The implementation ran two motors. You specify steps/sec, pattern, and step count for each motor. The pattern can be forward/backward using steps/half-steps. The motors run until the step count is reached, then the PIO throws an interrupt.
  - DMA and PIO based NTSC video generator which used zero cpu cycles to refresh the screen.
    there are 255x200 1-bit black/white and 4-bit gray scale versions. One use of this was to simulate diffusion-limited aggregation.
  - A scheme to extract the dominant frequencies (called formants) of vowels in speech. This is mathematically involved (three FFTs), but runs in real-time on the two cores. The code uses two interesting approximations. The first is the alpha-max, beta-min algorithm to speed up squareroot of sum-of-squares. It is accurate to within 2%. The second is an approximation of log base two from Generation of Products and Quotients Using Approximate Binary Logarithms for Digital Filtering Applications, IEEE Transactions on Computers 1970 vol.19 Issue No.02. It is accurate to within 0.02 log units.
  - There are also some tests for ISR speed, IIR fliter speed, and FIR speed. The FIRs are just fast enough to do Head Related Transfer Functions in real-time.
- Future:
  - packet radio or packet IR
  - adaptive filters
  - head related transfer function
- Obsolete ece4760 page -- for reference only!
Cyclone5 FPGA/SoC
- HPS Lattice-Boltzmann fluid flow solver aimed at the FPGA, but currently running on HPS side. The current code uses 20-bit fixed point in s1x18 format (sign-bit, one bit to left of binary-point, 18 to right). This format fits nicely into the M10k memory blocks which can be formatted as 512, 20-bit words.
- Quartus 18.1 ports. We need to move from Quartus 15 to 18.1 for continued support. This activity is aimed at checking compatability (and cleaning up examples). The only compatability issue so far is a change in a Qsys module of the VGA control system. The change invalidates two obsolete i/o devices, which happen to be set as defaults in Quartus 15. The fix is to use pull down device menus to choose valid devices.
- Using Verilog to understand the rp2040 PIO processors. The Pi rp2040 microcontroller has 8, single cycle, deterministic, Programmable i/o blocks (PIO). Each PIO is programmed using a 9-instruction assembly language. We have been speculationg about how you make a single-cycle machine (including jump). This project attemptes to answer that, without implementing the entire PIO architercture.
- ece5760 page
Analog
- Antialiasing.
  Build your own or buy a module.
- Analog computer
Math
- Lattice-Boltzmann
- Cellular Automata
- Computer graphics
- Cepstral analysis of speech