Hundreds of different simple SDR receivers have been designed around Dan Tayloe's Quadrature Sampling Detector or QSD.   Mine add nothing to the state of the art, and in fact subtract things, as I like minimalist solutions and the QSD is right in that sweet spot.   Following the evolution of Tayloe's design I delete the resistors in series with the sample lines for instance.   And to make the cheapest possible receiver that can be built wtihout SMT soldering skills,   I even deleted the faster 3253 mux/demux IC and substituted a lowly 4052.   It tops out at about 10 MHz but can be soldered in with a Weller gun if need be.    But the open question always was - How good is the QSD anyhow,  especially after being given the "Mad Man Muntz" treatment?

To that end, here is some data.    My test receiver is a very simple QSD, identical to that in the uSDX except there is no input filtering and the mux is a CD4052 rather than a 74CBT3253 (the 4052 has higher on resistance and its switching speed limits it to about 10 MHz.   Thus a 3253 should provide even better performance).      The outputs of the opamp (which is powered from 5VDC) are fed to a PC soundcard in parallel with the PICO ADC inputs (StarTek USB Audio with a transformer type ground loop isolator).

With this test receiver and a calibrated signal generator, the measured MDS is -122dBm in 500 Hz bandwidth.   This is the standard used by ARRL, RSGB, Rob Sherwood, etc.   This means the basic simple QSD has more than enough sensitivity to hear any signal that is above the noise floor in even the quietest locations (ref:   https://i.imgur.com/MKcaEbz.png).   This gives a useful benchmark for QSD receiver perfomance using the hi-fi soundcard and PC software DSP.

Since the op amp output voltage with -122dBm applied is very low,  I set the generator for 50uV (-73dBM) which is the standard RF level used to calibrate S meters (S9=50uV).   With 50uV input to the QSD the op amp output voltage was measured at 12mV RMS or 34 mV peak-to-peak, which is equal to -25 dBm.   I used the 25kHz lowpass input filter on the scope for all measurements to get rid of unrelated noise.

In the uSDX, Guido powered the 4562 opamp from 12 volts and I remember a comment that this was required to develop enough output voltage.   So I wanted to compare the output voltage with the opamp powered from both 5 and 12 VDC , and found no measureable difference at all.   The opamp output voltage was 12 mV  with an S9 signal input regardless of supply voltage (5 or 12VDC).

To validate this finding, without changing any of the test conditions I substituted one of my own uSDX boards for the Pico test receiver.   This board is a different layout but the receiver is identical to the published uSDX schematic including 12volt power to the opamp and again, no input filter was used.    The measured output with the -73dBm signal input was again 12 mV RMS but with quite a bit more noise (which may be due to layout, CPU differences, the OLED display, etc).

So the bottom line is:  the simple QSD receiver will convert a -73 dBm RF signal to a -25 dBm IQ baseband signal,  a gain of 48 dB.  Very impressive for two cheap ICs!

Comparison vs. uSDX:

The ATMega328 has a 10-bit resolution with +/-2LSB accuracy, the RP2040 ADC has been described as "woefully underspecified" but has 12 bits of resolution and 9 ENOB.   It should be noted that "Raspberry Pi has confirmed it is investigating reports of a design flaw in the analog to digital converter (ADC) on the Raspberry Pi Pico and other RP2040-based microcontroller boards, which manifests itself as measurable spikes in differential non-linearity (DNL)."   An explanation of the DNL issue can be found here: https://pico-adc.markomo.me/INL-DNL/#why-does-the-dnl-spike For simplicity let's just use raw ADC counts of 1024 for the ATMega328 vs 4096 for the Pico.   Both are biased to 50% so half these counts are used for positive and half for negative peak voltages.

Using Klaus's number of  800uV/ADC step (1.67V/2048),  this means a 16mV peak voltage corresponding to an S9 signal will be about 21 counts of the ADC.   From Guido's description we also know that "The ADC inputs are low-pass filtered (-40dB/decade roll-off at 1.5kHz cut-off) to prevent aliasing and input are biased with a 1.1V analog reference voltage to obtain additional sensitivity and dynamic range."   Reducing the AREF voltage (from the ~3.2V of the Pico) has the effect of increasing the ADC count for a given input (with the tradeoff of less dynamic range, of course, which is part of the uSDX gain control scheme).      The Pico doesn't offer the ability to change the ADC reference on the fly like this but an external reference voltage can be used if desired.   In any case, the ADC resolution of the ATMega328 with Vref set to 1.1 volts results in ADC sensitivity of .55v/512 or about 1mV per step - virtually the same as the Pico with its higher ADC resolution but also higher AREF (1.67/2048=.815mV/step).

For reference, with my test receiver to get 80mV P-P out of the opamps takes -67dBM input.   For 1 V P-P it takes -43dBm and 3 V P-P requires -31 dBm.    These measurements are consistent with the 48dB gain figure.