HFRX Prototype

HFRX Assembled


This is a test prototype for experimenting with a Quadrature Sampling Detector. It includes an Si570 I2C-tunable oscillator (0-160MHz), a divide-by-4 sequencer, a 1:4 analog mux and a low noise amplifier. A 50-ohm SMA connector is provided for RF input and an 8-pin 0.1" header provides power, I2C and differential analog I and Q output.



The Quadrature Sampling Detector (QSD) is a unique approach to signal downconversion which uses switched-capacitor techniques to simultaneously mix and filter a high-frequency signal down to baseband. I was first introduced to this concept by Dan Tayloe during a talk he gave at the Ft. Tuthill Hamfest several years ago. Since then, variations on the design have appeared in many projects popular throughout the amateur radio community, most notably the SoftRock series of HF receivers.

This circuit borrows heavily from many prior designs, in particular in its use of the Si570 I2C-tunable oscillator, the LV3253 analog mux and the LT6232 op-amp. One notable difference is that the baseband gain stage preserves the differential nature of the I and Q signals instead of combining them into unipolar signals. The reason for this choice will become clear below.

Overall System Application

Because the QSD generates as output a complex I and Q analog signal, further processing at baseband is required to recover usable audio or data. In the case of Amplitude Modulation (AM) a simple magnitude detector may be applied to demodulate the audio. Single Sideband (SSB) usually uses some variation on the Weaver demodulator and Frequency Modulation (FM) can be processed in the phase domain. While Tayloe's original design used analog circuits to implement the audio demodulation, more recent systems including the SoftRock radios feed the raw I and Q signals to a PC stereo audio input for processing in the digital domain. This provides a great deal of flexibility in the demodulation processing, including the ability to handle various digital modes.

While most current applications of the QSD rely on "big iron" PCs to do the DSP required for demodulation, the latest generation of embedded processors showing up on the market such as the ARM Cortex M4 have DSP capabilities that are sufficient to the task with a much smaller footprint in size, weight and power. On-board peripherals for many of these tiny MCUs include decent A/D converters that can digitize the analog I and Q signals for subsequent digital processing, and many provide either on-board DACs or peripheral interfaces sufficient for outputting high-quality audio.

The approach envisioned here is to use an STM32F373CCT6 MCU to do all baseband processing in the digital domain, outputting the demodulated audio via the on-chip I2S interface to a good quality stereo audio DAC. This MCU includes a set of 16-bit ADCs which can operate at sufficiently high rates to sample the baseband I and Q for the required processing. To maximize noise immunity in the ADC, the differential input mode will be used which requires that the gain stage on this prototype board not sum the opposing phases into a unipolar signal.

Prototype System

HFRX Prototype System

The prototype system consists of the HFRX board, an STM32F373 Breakout board, an I2S DAC PMOD, an Adafruit OLED board and a rotary encoder all mounted on perfboard and interconnected with wire-wrap. Two of the 'F373 SDADCs are configured to sample the differential I and Q signals at 18.75kHz (the highest common sample rate with the on-chip I2S port) and all the DSP takes place at this sample rate. After demodulation the processed audio is buffered into the I2S port for transmission to the audio DAC. The OLED display and rotary encoder provide a simple UI for controlling operating parameters. Presently, this system is currently capable of receiving AM over the range of 870kHz - 40MHz with realtime tuning of the LO and BFO frequencies. Further development will allow selection of baseband filter bandwidth, modulation, audio gain, etc. Additional refinements are envisioned, such as improved fine tuning and audio muting to eliminate pops during coarse tuning.

Lessons Learned

Starting with the prototype system built in 2014 I had a fair amount of initial success. Using only an AM demodulator I was able to confirm operation and the limits (both high and low) of sensitivity. As noted above, a smoothly ramped audio mute was added to prevent loud pops when tuning the LO.

Once these issues were addressed the major downside of the QSD quickly revealed itself - reception of images at odd harmonics of the Local Oscillator. When tuned to 10MHz, the system will also pick up weak images of signals at 30MHz, 50MHz, 70MHz, etc. Since this is a weakness of the QSD itself the only way to eliminate the images is to add preselect filters. I put the system away for a year and moved on to other things.

RF Filters

RF Filters

In June 2015 I became curious enough to build some filters. Knowing that only odd harmonics would cause problems, I designed six low-pass filters to cover the full range of the receiver, with corners at 1.6, 3.2, 6.3, 11.5, 22 and 40 MHz. A 5th-order Chebyshev shunt design was used and L/C values chosen from the 5% range. I found that the 0603 SMD inductors available today perform nicely and I was seeing excellent matches to the design characteristics.


Putting these filters inline between the antenna and HFRX prototype system allowed me to further test performance without seeing spurious response from higher up the spectrum (originally I was getting significant interference from broadcast FM stations) and work on addition refinements. First up was to add Single Sideband (SSB) demodulation. Using and all-pass IIR implementation of a Hilbert Transform that'd I'd first developed for an electronic music frequency shifter, I was abled to get rudimentary upper and lower sideband demodulation working.

Received Signal Strength Indicator

The original AM demodulator works by computing the square root of the sum of squares of the I and Q channel, then applying a DC block to remove the carrier component. By low-pass filtering the raw sum of squares, then converting to dB it's possible to create an estimate of input signal strength. Calibrating vs a known source (my own RF signal generator with accurate padding) it was possible to determine that the receiver saturates at about -27dBm (roughly 10mVpp) which corresponds nicely to the 1Vpp input range of the STM32F373 SDADCs along with the 100x gain of the differential amplifiers in the QSD. A discernable signal was heard when the power was reduced to -106dBm for an overall dynamic range of about 80dB.

Audio Filters

For improved audio quality it's helpful to filter out high frequencies beyond those necessary to understand a transmission. For CW this can mean a filter cutoff as low as a few hundered Hertz while SSB phone might be more like 4kHz and broadcast AM might go as high as 8kHz. Between the BFO and demodulator I added a pair of 6th-order IIR filters implemented as a chain of second-order sections. I build coefficient sets to support six bandwidths of 8kHz, 6kHz, 4kHz, 2kHz, 1kHz and 500Hz.


With 80dB of dynamic range I found myself constantly tweaking the headphone amplifier gain as I switched from a super-powerful nearby AM broadcast station vs the evening SSB nets on 20M. Adding an AGC corrected this. The design used here is a log-domain AGC with a settling time of about 0.5s. The AGC sits between the Audio Filters and the Demodulator. RSSI was revised to report the AGC gain setting instead of just the measured signal power.

Aliasing Troubles

After adding the AGC it became clear that there were some issues with SSB demodulation. I'd frequently hear the tell-tale quack-quack of an inverted sideband and shifting to the opposite mode wouldn't fix it. Running tests on the audio chain in a command-line test bench with synthesized signals showed no particular problems, so it must have been something upstream in the system. Revisiting the analog signal flow showed a problem - the signals from the QSD to the SDADC inputs of the STM32F373 were extremely wide bandwith and were likely aliasing. Recomputing the QSD sampling capacitor values suggested that the initial 0.01uF selection was likely 25x wider than necessary, and the op-amps were also much wider than needed. Substituting in larger capacitors to reduce the bandwidth to something nearer the 9kHz nyquist frequency of the system produced remarkable improvements in birdies, audio quality and allowed the USB / LSB selection to operate as desired.


RF Filters

Audio processing software on the STM32F373 with all the features noted above is working well. CPU load with muting, BFO, filters, AGC and SSB demodulation runs about 40% of total cycles available providing plenty of headroom for other features if needed. Only about 25% of the total flash ROM is used.

Next Steps

It's clear that for a complete system with decent performance there are a few improvements needed to the current architecture:

Design Resources


Return to Radio page.

Last Updated
Comments to:
Eric Brombaugh

Valid HTML 4.01 Transitional