Super Low Distortion Ultra Pure Audio Oscillators Revisited

[JR.]

Nice, you are already tens of dB better than I could even hear.... When dialing in the TS-1 I could just about hear distortion components down 50-60 dB. Which makes you twice as good (in dB terms, that is more like 1000x better in real linear world).

I am impressed by how smart DSP can be applied to extract resolution way down in the dirt.

I killed a bunch of brain cells worrying about the original TS-1 design, I suspect a next generation TS-1D would kill the brain cells left.

JR

[mediatechnology]

The converter and output stage for a NCO/DDS is still the weak link.

With all the hardware and advanced math shown in the ADI document to generate a sinewave the best performance they can come up with is only equal to a high quality Soundcard or outboard converter.

The analog oscillator is so much better from a purity standpoint it can be used to quantify converters.
Converter linearity testing how many of the oscillators in the OP came to be.

This is the distortion signature of the Focusrite 2i2 A/D at 0 dBu.
The red trace is the oscillator seen through the notch output with +40 dB gain and its virtually distortionless.
The blue trace is the 2i2's A/D distortion signature.



I couldn't have quantified the 2i2's A/D using any DAC I have.

[mediatechnology]

Here's the schematic of the notch filter I've been using borrowed from "Janasek."


"Nacho" 1kHz Notch Filter

"Nacho's" insertion loss at 2 kHz is about -9dB; at 3 kHz the loss is about -5 dB.

[mediatechnology]

"Nacho" Notch Filter frequency response sweep and insertion loss.


"Nacho" Notch Filter Frequency Response

I measured the insertion loss with a function generator, frequency counter and the Fluke 8050 voltmeter to avoid measurement error using AudioTester's cursors.


"Nacho" Notch Filter Insertion Loss

My next project is a spreadsheet that will allow me to plug in values for the HD2, HD3, HD4 etc. harmonic levels, power sum them and then display the overall THD level in dBc and percent.
The calculator will take into account the reference level, post-filter gain and filter insertion loss.

[mediatechnology]

"ULDO" versus the Focusrite 2i2 D/A

The following shows the spectral purity of the all-analog approach for generating ultra-pure sine waves.
The Focusrite 2i2 admitedly isn't a top-of-the line D/A but the THD, at 0 dBu, is respectable at 0.00068%
"ULDO" however has an order of magnitude advantage in THD and is much more pure at <0.000028%.
The 0.000028% figure assumes that there is distortion in the noise floor at approximately -131 dBu. (-140 dBu +9 dB filter loss at HD2.)

Note that the displayed Y-axis should have 40 dB subtracted from it due to the post-filter gain.
-100 dBu is actually -140 dBu.
The 1 kHz spur is the notch filter residual.


ULDO 0 dBu 40dB Notch Output


Focusrite 2i2 0 dBu 40dB Notch Output

The Analog Devices article puts spurs at -120 dBc as state-of-the-art for a DDS oscillator with performance limited by the D/A.
Mathematically-generating pure sinewaves isn't the limitation - its the converter.

[terkio]

Insertion loss.
Theoretical values calculated with Spice.
10nF and 16K as " Nacho 1KHz notch filter".

Notch 994.7 Hz
100 Hz -0.66 dB
200 Hz -2.31 dB
500 Hz -9.16 dB
2KHz -9.03 dB
3KHz -5.08 dB
4KHz -3.27 dB
5KHz -2.27 dB
6KHz -1.66 dB
7KHz -1.26 dB
8KHz -0.99 dB
9KHz -0.79 dB
10KHz -0.65 dB
20KHz -0.17 dB

[mediatechnology]

Thanks for running the sim!
Looks like it's pretty close.

On the oscillator prototype I decided to try 15K8 1% and the three I grabbed at random came in at 999 Hz.
I got lucky.
I'm using 1% caps and they may just be a tad high in value.

I have a spreadsheet I wrote to plug in the distortion levels from the FFT, subtract the gain, subtract the carrier level, add the filter insert loss, compute the ratio and then power sum HD2 to HD10 to obtain the THD.
THD is displayed as percent, parts per million and dBc.

This is the FFT of the DCAO-II headphone amplifier at 500 mW into 30Ω.


FFT of the DCAO-II headphone amplifier at 500 mW into 30Ω.
Red is notch filter out. Blue is an attenuated sample of the output.
The DCAO-II and Uldo both have distortion that is less than the residual of the Focusrite 2i2's A/D.

Plugging the FFT data into the spreadsheet allows THD to be calculated.


Spreadsheet calculates the THD of the DCAO-II to be 0.00023% based on FFT measurement of the notch filter output.

[mediatechnology]

I followed my own advice and re-routed the ground lines from the LM78/7912 sub-regulators.

This reduced H2.

This is an FFT of ULDO producing +20 dBu into an open circuit loaded only by the notch filter.





The THD of the Ultra Low Distortion Oscillator is -143.94 dBc; 0.00000635% or <0.1 ppm.

[JR.]

I don't know if this offers any insights but I vaguely recall an old AES Journal paper written by Cherry about the impact of PCB layout on audio amplifier THD. As I recall it mostly about the PCB layout around the power transistor output stage. Certainly dealing with larger currents but perhaps similar mechanisms.

JR

[mediatechnology]

I think I have the Cherry paper.

For unbalanced outputs moving the output return current to a better location reduced its modulation of the LM78L12/79L12 ground references.
I drew what I wanted on the schematic but didn't route it that way on the Protoboard.

What I did find is that most of the HD2 was due to 5532 load current when driving balanced loads.
Balanced loads don't return current to ground - they feed current into the opposing output - so I turned my attention to the 5532's supply pins.
A 470 uF capacitor connected differentially at the 5532's supply pins (from Vcc to Vee) virtually eliminates HD2.
Some of this is likely due to parasitic resistance in the Protoboard but I'm going to put a big honkin' cap at the device pins anyway.

At less than +16 dBu balanced into 600Ω the distortion is <-140 dBc.
+18 dBu balanced 600Ω load is -138.9 dBc (all HD3). This is 63 mW into the 600Ω load - the 5532 is delivering 126 mW. (+24 dBu into 1200Ω. There are 2X 301Ω build-outs.)
+20 dBu balanced 600Ω load is -134.2 dBc (all HD3). 100 mW into 600Ω; 5532 200 mW.
+26 dBu balanced into an open circuit is -142.9 dBu. (About even HD2/HD3.)

I think the remaining distortion is mostly related to 5532 output stage linearity.
It's good enough that I don't want to add a buffer.

One thing that's really good about this oscillator is that the output level is set by a DC current.
The useful range is about -12 dBu to +26 dBu.
I found a really nice old 20 turn wirewound pot that I'm using to set level.
It's pretty easy to dial level in to 0.01 dB and the tempco is pretty low.
I see level drift of about +/-0.02 dB with typical room temperature variations.

Here's the latest schematic:


Updated schematic 1/8/21

Large: https://proaudiodesignforum.com/images/ ... ropped.png