Patent US9590648B2 - 28 bits ENOB DAC -> 168 dB SNR

[analityk]

Hi folks.
I do not believe in this parameters can be obtained in room temperature equipment.
Author claim due to his patent he achieved 40 nV broadband noise in 10 V RMS output signal in bandwidth between 20Hz-20kHz in his DAC.
In this paten is explanation how it works but it seems to be math trick more than real dynamic range.
OFC there is no equipment ready to measure so low noise and this awesome linearity.

This photo and parameters are an effect of linkedin serious buissness discussion.

I just do not believe, I will ask you for your calculation and opinions.

In my opinion even this component density is to big to be ready reach so low noises.

Sincerely.

[dietert1]

This system is similar to an old Philips method used in the Philips DSS audio system. Their audio DAC also had two DACs and they were wired to produce mantissa and exponent of a high resolution audio signal. Exponent means a gain level. When you test such audio systems and you feed a full scale signal, noise will be higher than when you feed silence. The subjective dynamic range is measured as full scale output in comparison to noise with silence, so it can be very good.
The cited invention explains that they turn off the high path when it isn't needed, so something similar happens. See patent text column 2 lines 65ff.

Regards, Dieter

[analityk]

Och, ok. I mean I got it. This is "subjective" 28 ENOB. During full scale signals noise floor is much higher than during silence.
Some magic tricks is still how they want to produce high voltage signals without noise.
Does this DAC change gain dynamically? Any way, I've asked author about total integrals nonlinearity, but he claim their DAC "is linear". Does changing gain not introduce some distortion in signal?

[dietert1]

Yes, changing the gain during normal operation may create glitches and one tries to reduce the rate of gain changes to a minimum. So one needs kind of a level meter, maybe with FFT so the delay will depend on the spectral properties of the actual audio signal.
In the patent they use the term "passive adder" multiple times. Maybe they don't change the gain, but just mute the contribution of the high path. Once more a good chance to add glitches.

Regards, Dieter

[MiDi]

THD+N at 1kHz (22kHz BW): 0.0002% at +22dBu
Quite good, but not top notch.

[TimFox]

Before really high resolution ADCs were fast enough for digital imaging systems (e.g., x-ray detectors for CAT scanners), it was common to do a flash conversion of the most-significant four bits, and use that to control the gain before a 12-bit or 16-bit digitizer.
Of course, the x-ray signal being measured was approximately logarithmic in the path length (areal density), so the result was a good match to the requirement.