Close
0%
0%

HF Receiver on ADC and FPGA

I brought up a commercial Zynq board, captured a real ADC signal, and built a working FPGA receiver path.

Similar projects worth following
A working FPGA-based HF receiver platform built from a SmartZynq SP2 and an AD9248 dual ADC board. The SmartZynq now boots Xillinux with HDMI, Ethernet and SSH; the FPGA stream path has been tested through Xillybus; a simulated 64 MS/s RF source, tunable DDC, sharp channel filter, USB/LSB demodulator and FT8-like test source are working in the current design. Physical AD9248 channel-A capture has also been tested at 64 MS/s, characterizing the analogue front end. Now the next step is to scale the verified receiver channel into multiple independently tuned digital-mode receivers.

OVERVIEW: one small station, many listeners

two-board setup

The AD9248 board (on the left) is the acquisition end: 

  • antenna input conditioning 
  • and high-speed conversion. 

https://www.analog.com/media/en/technical-documentation/data-sheets/ad9248.pdf


The SmartZynq SP2 (on the right) is the processing end: 

  • it receives the sample stream, 
  • runs the digital signal processing pipeline in programmable logic, 
  • provides the operating system and network interface.

http://www.hellofpga.com/index.php/2024/01/21/smart-zynq-sp2/

The firmware turns a single digitized RF high speed stream into a set of configurable receive channels. 

Each channel has its own digital oscillator, filter, decimator, and output stream. 

This makes the hardware reusable: choose the frequencies in configuration instead of rewiring a receiver.

The intended first-use flow is:

1. Flash the supplied SD-card image or firmware bundle.
2. Connect the ADC board to the SmartZynq SP2 and attach the antenna.
3. Connect power and Ethernet.
4. Select a preset—such as multi-band FT8, WSPR monitoring, or a custom channel plan.


Channels are software, not hardware

The FPGA performs the repetitive high-rate work close to the ADC: digital down-conversion, filtering, decimation and demodulation. The processor side packages lower-rate IQ or audio streams for a local decoder, a network client, or logging software.

Example channel plans:

PresetExample allocation
FT8 watcherOne receive slice per selected amateur band
WSPR beacon monitorPersistent WSPR slices plus spare experimental channels
Mixed digital benchFT8, WSPR, CW/skimmer, and a general-purpose IQ stream

Actual simultaneous channel count, usable coverage, and dynamic range will be documented from measured firmware builds. 

Input filtering and antenna protection remain important, especially with strong nearby transmitters.


Why these two boards?

The AD9248 provides fast dual-channel conversion, while the SmartZynq SP2 pairs FPGA fabric—ideal for many parallel DDCs—with an embedded processor that can boot, configure the receiver, and serve results over Ethernet. Together they move the difficult hardware work into a repeatable two-board assembly and leave experimentation to firmware.


Block diagram

The antenna enters the ADC board, where it is protected, filtered, and matched before conversion. 

Sample data crosses to the SmartZynq SP2, whose FPGA creates independent digital channels. 

The SoC then exposes those channels as IQ, audio, or decoded/spot data over the network.

Project status (see the logs for details)

  • The vendor Xillinux design rebuilt in Vivado 2026.1
  • A 64 MS/s virtual RF source through the FPGA DDC
  • USB/LSB audio generated in PL and played on an Ubuntu PC
  • PS-programmable 79-symbol FT8 test waveform
  • Why the AD9248 output needed offset-binary conversion
  • From one verified receiver channel to eight planned channels
  • Architecture and project presentation are in place. 

Next milestones: 

  • test multi channel
  • run some FT8 decoders on PS

The ADC, DDC and demod interface

Using this interface you can set:

  • source: an internal FT8 generator or the external Antenna
  • data processing: raw ADC or DDC I/Q or demodulated SSB (USB or LSB)
  • real time view and spectrum view

Receiving an external signal (1KHz USB)

ADC - DDC - CIC FIR filtering - SSB demodulator Vivado Block design

This is the top-level Vivado block design. It connects the Zynq processing system, the external ADC, the virtual FT8 source, the FPGA receiver chain, and the Xillybus DMA interfaces.

PS System is the vendor SmartZynq platform: the Zynq PS, DDR and AXI infrastructure, XillyVGA/HDMI output, Ethernet/RGMII, UART, GPIO, and clock/reset generation. It runs Xillinux and provides the AXI connections used by the FPGA logic.

External ADC Input receives the parallel AD9248 channel-A data and over-range signal. It runs in the ADC clock domain,...

Read more »

  • 1 × HelloFPGA SmartZynq SP2 V1.3B The Zynq 7020 board
  • 1 × AD9248-65 dual ADC module The 14 bits 64MHz ADC

  • FT8 decoding from the FPGA's simulated audio stream

    Guido - IW5ALZ • 5 days ago • 0 comments

    The receiver can now decode an FT8 transmission generated entirely inside the SmartZynq FPGA. This provides a repeatable test of the complete path from the programmable tone generator, through the DDC and USB/LSB demodulator, into the Xillybus audio stream and the Cortex-A9 decoder.

    The PS uploads a 79-symbol FT8 frame to the FPGA through /dev/xillybus_mem_8. The default frame is a valid FT8 transmission:

    G0XYZ K1ABC FN43

    The FPGA emits the frame as a 1 kHz audio signal. USB audio is sent as 8 kHz, 16-bit little-endian PCM over /dev/xillybus_read_8. The Xillybus endpoint transfers one byte at a time: low PCM byte followed by high PCM byte.

    Setting up the simulated receiver

    Set the receiver source, demodulator, DDC frequency, and simulated carrier once after boot or FPGA configuration:

    python3 /root/receiver_control.py \
      --source simulated \
      --mode usb \
      --tune 1000000 \
      --sim-frequency 1001000

    The 1 kHz difference between the simulated carrier and DDC tune becomes the FT8 audio frequency. Uploading the default tone vector starts or restarts the 15-second repeating FPGA transmission:

    python3 /root/ft8_control.py

    ft8_control.py writes the packed 79 tones to the FPGA, verifies every byte, commits the new frame, and waits for the FPGA acknowledgement.

    Slot timing matters

    The live decoder opens /dev/xillybus_read_8 at a UTC 15-second boundary and captures 120,000 PCM16 samples. The first seven FT8 symbols form a Costas synchronization block, so the generator must begin inside the captured slot. Starting it just before the boundary cuts off that first block and prevents synchronization.

    For this test, the tone upload was scheduled 0.75 seconds after the next UTC boundary while the stream decoder waited for the same boundary:

    cd /root/ft8_decoder
    make stream
    
    (
    python3 - <<'PY'
    import subprocess, time
    target = (int(time.time()) // 15 + 1) * 15 + 0.75
    time.sleep(max(0.0, target - time.time()))
    subprocess.run(["python3", "/root/ft8_control.py"], check=True)
    PY
    ) &
    
    build/a9-stream/ft8-subtract-stream-a9 --slots 1 --base-mhz 1.001
    wait

    The Cortex-A9 produced the expected decode:

    snr=+20 dt=+1.14s freq=1000.0Hz text=G0XYZ K1ABC FN43

    The stream runner uses the same three-pass residual-subtraction decoder as the WAV test program. It is built with an 80-candidate search limit, which also allows the receiver's 8 kHz test recording to recover all 27 known messages.

    What this proves

    This test validates the FPGA tone upload, simulated RF source, DDC, SSB audio path, PCM byte order, Xillybus transport, UTC-aligned capture, and native FT8 decoder as one chain. The generator is continuous-phase 8-FSK and does not yet apply FT8 Gaussian frequency shaping, so it is a deterministic integration test rather than a substitute for a weak-signal off-air test.

  • USB/LSB audio generated in PL and played on an Ubuntu PC

    Guido - IW5ALZ • 10/22/2026 at 09:33 • 0 comments

    The receiver now produces demodulated audio directly in the SmartZynq FPGA fabric.

    After the 64 MS/s input is digitally mixed and filtered down to an 8 ksample/s complex I/Q stream, a 127-tap Hilbert-filter demodulator generates both USB and LSB audio. The Linux processor does not perform the SSB demodulation: it only configures the receiver and reads the resulting PCM stream through Xillybus.

    The FPGA provides continuous signed 16-bit, mono, 8 kHz PCM through:

    /dev/xillybus_read_8

    The stream crosses from PL to PS through an asynchronous FIFO. This is separate from the 4096-sample diagnostic snapshots used by the waveform/spectrum monitor, so audio continues while the receiver is running.

    I first tested the path with the internal simulated RF source. The simulator carrier was placed 1 kHz above the DDC tuning frequency, so USB produced a 1 kHz audio tone while LSB rejected it. Moving the carrier 1 kHz below the tuned frequency reverses the result, which checks the sideband convention and demodulator sign.

    The board has no audio DAC yet, but Xillinux can stream the PL-generated PCM data over SSH to an Ubuntu PC. After configuring the receiver:

    ssh root@192.168.1.126 \  'python3 /root/receiver_control.py \  --source simulated --mode usb \  --tune 1000000 --sim-frequency 1001000'

    I can listen on the Ubuntu PC with:

    ssh -T root@192.168.1.126 \  'cat /dev/xillybus_read_8' |
    aplay -t raw -f S16_LE -c 1 -r 8000

    This confirms the complete path:

    RF source → ADC/simulator → DDC → sharp channel filter          → USB/LSB demodulator in PL → Xillybus DMA          → Xillinux → SSH → Ubuntu sound card

  • Testing the external ADC input on SmartZynq SP2

    Guido - IW5ALZ • 10/17/2026 at 19:54 • 0 comments

    After validating the FPGA ADC simulator, I connected the AD9248 board to the SmartZynq SP2’s J6 header and tested physical channel A.

    The FPGA supplies an 8 MHz sampling clock to channel A, with channel B’s clock held low. Each acquisition captures 4096 samples—a 512 µs snapshot—and transfers them through Xillybus to Xillinux. A Python 3/Matplotlib application running on the Zynq PS displays the waveform, spectrum and signal levels, locally on HDMI or remotely through VNC.

    A PS-controlled selector switches between simulated and external input without rebuilding the bitstream. Selection remains fixed throughout each capture.

    For the external test, I applied a 1 MHz sinusoid at approximately 1 Vpp to the ADC board, which has approximately 14 dB input attenuation. The first waveform looked severely distorted. Reinterpreting the samples revealed that the board outputs offset binary, while our capture path expected two’s complement.

    Software conversion recovered a clean 1 MHz sine at approximately −15 dBFS, with no samples reaching the digital rails. The FPGA input logic was then updated to perform this conversion directly by inverting bit 13 before sign extension. The rebuilt design passed simulation and timing checks; its final live hardware verification is the next step.

  • SmartZynq SP2 bring-up: factory backup and Xillinux tests

    Guido - IW5ALZ • 10/07/2026 at 18:33 • 0 comments

    My HelloFPGA SmartZynq SP2 V1.3B board arrived and ran its factory demo correctly. Before replacing anything, I backed up its QSPI flash and tested the board with Xillinux.

    Preserving the factory demo

    Using XSDB 2026.1 and the onboard JTAG interface, I halted both Cortex-A9 cores and checked the QSPI controller. Linear read mode was enabled, and the flash contained the expected Zynq boot-header markers.

    I read the entire 16 MiB QSPI flash twice:

    mrd -size w -bin -file smartzynq_factory_qspi_A.bin 0xFC000000 4194304
    mrd -size w -bin -file smartzynq_factory_qspi_B.bin 0xFC000000 4194304

    Both files matched byte for byte, with SHA-256:

    a06a3866508a02b8bc8c3cb9d9dd08966a8cc12951b2e228092c80f1b7a1f2b4

    Inspection revealed three unencrypted partitions:

    Size
    98,312 bytes
    4,045,568 bytes
    81,928 bytes

    The remaining flash space was erased (0xFF). The PS application contains Xilinx USB virtual-disk strings and UART1(PS):HELLOFPGA.COM, consistent with the factory demo.

    This backup preserves the compiled firmware, not the original HDL or C sources. Restoration has not yet been tested.

    Booting Xillinux

    I prepared a microSD card with Xillinux and the HelloFPGA SmartZynq SP2 V1.3 reference bundle. Its boot partition contains:

    uImage
    boot.bin
    devicetree.dtb
    xillydemo.bit

    The supplied prebuilt bitstream booted successfully on my V1.3B board. Linux displayed its terminal over HDMI, and Ethernet networking worked.

    After setting a root password with passwd, I connected from my computer using:

    ssh root@<board-IP>

    Testing PS and PL

    The reference design exposes LED2 as /sys/class/leds/mmc_led. I temporarily disabled its SD activity trigger, switched its brightness between 0 and 255, then restored the mmc0 trigger. LED2 responded correctly, confirming control from Linux through PS GPIO and PL routing.

    The Xillybus devices were also present:

    /dev/xillybus_mem_8
    /dev/xillybus_read_8
    /dev/xillybus_write_8
    /dev/xillybus_read_32
    /dev/xillybus_write_32

    Both stream pairs use FIFO loopbacks in the FPGA. Two tests passed:

    • An 8-bit text transfer returned the expected message.
    • A 32-bit transfer returned 4,096 bytes of random data, verified byte for byte with cmp.

    These tests establish a working baseline for HDMI output, Ethernet, SSH, LED control, and bidirectional PS–PL data transfer.

View all 4 project logs

Enjoy this project?

Share

Discussions

Similar Projects

Does this project spark your interest?

Become a member to follow this project and never miss any updates