E1DA Cosmos ADCiso Review

[KSTR]

Audio Precision is not selling an ADC. It is selling a measurement system with all of its features built-in. Members like to generate results that are comparable to mine. Since I use high performance sine analyzer, then they need to know if they can get there with this box as is.

The HPSA (High Performance Signal Analyser) with its outstanding specs is only used for single sine test signals as there is only one notch filter. So the question of the "pure" ADC performance is fully legit.

For multitone the normal ADC path is used, and that path is very comparable in performance to the stripped down signal path in the Cosmos ADC with the ES9822pro, in other words, the Cosmos ADC is on par with AP for anything else than pure sine.

 

[pma]

Suppression of aliases at 192kHz sampling frequency. Input signal is 186kHz sine from analog generator, 1.44Vrms. ADC input range is 2.7Vrms.

 

[amirm]

The HPSA (High Performance Signal Analyser) with its outstanding specs is only used for single sine test signals as there is only one notch filter. So the question of the "pure" ADC performance is fully legit.

It is not "fully" legit. Many of the core tests are single frequency starting with FFT in the dashboard. THD+N vs level. THD+N vs frequency. THD+N vs power. And of course SINAD. Again, the goal is try to create comparable measurements to mine. For that you need to do well in these core tests.

 

[pma]

The key is in the autoranger, to compete in level sweeps. Distortion and noise at one selected range, at the rated level, would be similar, same or better than with the AP. Sweeps are not, for the noise reason.

However, ownership of the expensive test gear is not everything. And the ASR test set does not cover everything. To me, the goal is to find most revealing methods to examine the DUT to its limits.

 

[KSTR]

Suppression of aliases at 192kHz sampling frequency. Input signal is 186kHz sine from analog generator, 1.44Vrms. ADC input range is 2.7Vrms.

View attachment 366820

Which anti-aliasing filter did you use?
With "Linear Phase Apozing" I get perfect supression (down in the noise floor) of frequencies > fs/2 and and full scale input, at all sample rates up to 384kHz. Only very very close to fs/2 the alias needle is visible.

 

[pma]

I have made no filter selections. I installed the Cosmos driver according to its web page (I have win8.1 so the driver must be installed) and that was it. No filter selection.

Have you tried the input frequency close to Fs as I did?

 

[welwynnick]

In my early years I used play with lots of lab gear - oscilloscopes, sig gens, power meters and spectrum and network analysers.
I recall the latter had wide frequency and dynamic ranges, and I couldn't help wondering how well a modern signal analyser would fare as an audio analyser.
The vast majority don't go down below 9kHz; a few do reach 10 or 20Hz; but they're very expensive.
I hadn't realised, but there are actually quite a few dedicated audio analysers, with some familiar names, so I did a survey.
Typically, they have two analogue input/output channels that work over 10 to 80kHz, and many have options for digital I/O and additional channels.
There's a wide range of maximum voltage, and the I/O impedance is usually 100kohm/100 ohm.
The best ones I could find have a SINAD over 100dB, and they're all expensive (if they're even available at all).
The Rhode & Schwarz UPL models are also quite popular, but I couldn't find any specs (I guess it sits somewhere low down).
The Cosmos price includes the scaler so the input impedance is comparable, and the SINAD is Amir's stereo measurement (as they're all 2 channel).
As Audio Precision claims, nothing beats the APx555.
But I thought it would be useful to get some perspective on what E1DA and QuantAsylum had achieved, despite the best efforts of the big guns.

Manufacturer	Model	Channels	Bandwidth	Analogue I/O	Digital I/O	Price $	Spec SINAD	Typ SINAD
Audio Precision	APx555	2	200kHz	Y	Y	30,000	117	120
E1DA	Cosmos	2	190kHz	Y	N	298	117	117
Audio Precision	SYS2722	2	80kHz	Y	Y	26,000	112	115
Stanford Research Systems	SR1	2	200kHz	Y	Y	17,000	109	113
Agilent / HP	U8903A	2	100kHz	Y	O	(6000)	108	110
Audio Precision	APx525	2-4	80kHz	Y	Y	27,000	108	110
Keysight	U8903B	2-8	80kHz	Y	O	28,000	108	110
QuantAsylum	QA403	2	80kHz	Y	N	600	105	110
Rhode & Schwarz	UPV	2-8	250kHz	Y	O	12,000	105	110
Audio Precision	APx516	2	80kHz	Y	Y	6,000	109	109
PrismSound / Spectral	dScope III	2	200kHz	Y	Y	4,400	105	108
PrismSound / Spectral	dScope M1	2	90kHz	Y	Y	4,400	105	108
Rhode & Schwarz	UPP	2	80kHz	Y	Y	(7000)	100	105
QuantAsylum	QA401	2	80kHz	Y	N	450	101	101

 

[staticV3]

In my early years I used play with lots of lab gear - oscilloscopes, sig gens, power meters and spectrum and network analysers.
I recall the latter had wide frequency and dynamic ranges, and I couldn't help wondering how well a modern signal analyser would fare as an audio analyser.
The vast majority don't go down below 9kHz; a few do reach 10 or 20Hz; but they're very expensive.
I hadn't realised, but there are actually quite a few dedicated audio analysers, with some familiar names, so I did a survey.
Typically, they have two analogue input/output channels that work over 10 to 80kHz, and many have options for digital I/O and additional channels.
There's a wide range of maximum voltage, and the I/O impedance is usually 100kohm/100 ohm.
The best ones I could find have a SINAD over 100dB, and they're all expensive (if they're even available at all).
The Rhode & Schwarz UPL models are also quite popular, but I couldn't find any specs (I guess it sits somewhere low down).
The Cosmos price includes the scaler so the input impedance is comparable, and the SINAD is Amir's stereo measurement (as they're all 2 channel).
As Audio Precision claims, nothing beats the APx555.
But I thought it would be useful to get some perspective on what E1DA and QuantAsylum had achieved, despite the best efforts of the big guns.

Manufacturer	Model	Channels	Bandwidth	Analogue I/O	Digital I/O	Price $	Spec SINAD	Typ SINAD
Audio Precision	APx555	2	80kHz	Y	Y	30,000	117	120
E1DA	Cosmos	2	180kHz	N	N	298	117	117
Audio Precision	SYS2722	2	80kHz	Y	Y	26,000	112	115
Stanford Research Systems	SR1	2	200kHz	Y	Y	17,000	109	113
Agilent / HP	U8903A	2	100kHz	O	O	(6000)	108	110
Audio Precision	APx525	2-4	80kHz	Y	Y	27,000	108	110
Keysight	U8903B	2-8	80kHz	O	O	28,000	108	110
QuantAsylum	QA403	2	80kHz	N	N	600	105	110
Rhode & Schwarz	UPV	2-8	250kHz	O	O	12,000	105	110
Audio Precision	APx516	2	80kHz	Y	Y	6,000	109	109
PrismSound / Spectral	dScope III	2	200kHz	Y	Y	4,400	105	108
PrismSound / Spectral	dScope M1	2	90kHz	Y	Y	4,400	105	108
Rhode & Schwarz	UPP	2	80kHz	Y	Y	(7000)	100	105
QuantAsylum	QA401	2	80kHz	N	N	450	101	101

Nice comparison table!

One slight correction is that with a FW update, the Cosmos supports up to 768kHz Fs. That expands the BW to 384kHz (flat FR only with compensation file attached below).

 

[KSTR]

have made no filter selections. I installed the Cosmos driver according to its web page (I have win8.1 so the driver must be installed) and that was it. No filter selection.

For the filter selection you need to download and run "Tweak_Cosmos.rar".

I now tried to replicate your plot, running 188kHz -6dBFS, @192kHz SR, 10++ averages and get basically the same result. Initially I had used short FFT size which did hide the alias a bit more.

A bit worse than spec (which says around -100dB or better for most filters), but still OK in my book.

 

[KSTR]

One should not overrun the Cosmos with too sharp edges (dv/dt). Sine at 1Mhz near fullscale (which is admittedly brutal) gives this:

 

[Watire]

Is -6dbFS a good test signal level for mainstream benchmarking ? Or reaching the closer to clipping level is the best practice ?
If a device plays better numbers at -12dBFS is it valid ?
Can you read the absolute voltage output level of a device with the Cosmos capture from the get go ?

 

[GXAlan]

Is -6dbFS a good test signal level for mainstream benchmarking ? Or reaching the closer to clipping level is the best practice ?
If a device plays better numbers at -12dBFS is it valid ?
Can you read the absolute voltage output level of a device with the Cosmos capture from the get go ?

You have to calibrate the Cosmos to actual voltage. 10V nominal was 10.2V for 0 dBFS for me.

As close to clipping without intersample overs is the best.

 

[pma]

One should not overrun the Cosmos with too sharp edges (dv/dt). Sine at 1Mhz near fullscale (which is admittedly brutal) gives

Yes. I did try that yesterday as well, but did not post. Those are the reasons why I posted another thread in 2022:

Why ADC converters do not use proper input anti-alias filters?

Why ADC converters do not use proper anti-alias filters? As we all know, the Nyquist theorem states a necessary condition of proper sampling - input signal must not contain frequencies above Fs/2, half of sampling frequency. If this condition is violated, we are experiencing alias frequencies...

www.audiosciencereview.com

 

[restorer-john]

In my early years I used play with lots of lab gear - oscilloscopes, sig gens, power meters and spectrum and network analysers.
I recall the latter had wide frequency and dynamic ranges, and I couldn't help wondering how well a modern signal analyser would fare as an audio analyser.
The vast majority don't go down below 9kHz; a few do reach 10 or 20Hz; but they're very expensive.
I hadn't realised, but there are actually quite a few dedicated audio analysers, with some familiar names, so I did a survey.
Typically, they have two analogue input/output channels that work over 10 to 80kHz, and many have options for digital I/O and additional channels.
There's a wide range of maximum voltage, and the I/O impedance is usually 100kohm/100 ohm.
The best ones I could find have a SINAD over 100dB, and they're all expensive (if they're even available at all).
The Rhode & Schwarz UPL models are also quite popular, but I couldn't find any specs (I guess it sits somewhere low down).
The Cosmos price includes the scaler so the input impedance is comparable, and the SINAD is Amir's stereo measurement (as they're all 2 channel).
As Audio Precision claims, nothing beats the APx555.
But I thought it would be useful to get some perspective on what E1DA and QuantAsylum had achieved, despite the best efforts of the big guns.

Manufacturer	Model	Channels	Bandwidth	Analogue I/O	Digital I/O	Price $	Spec SINAD	Typ SINAD
Audio Precision	APx555	2	80kHz	Y	Y	30,000	117	120
E1DA	Cosmos	2	180kHz	Y	N	298	117	117
Audio Precision	SYS2722	2	80kHz	Y	Y	26,000	112	115
Stanford Research Systems	SR1	2	200kHz	Y	Y	17,000	109	113
Agilent / HP	U8903A	2	100kHz	Y	O	(6000)	108	110
Audio Precision	APx525	2-4	80kHz	Y	Y	27,000	108	110
Keysight	U8903B	2-8	80kHz	Y	O	28,000	108	110
QuantAsylum	QA403	2	80kHz	Y	N	600	105	110
Rhode & Schwarz	UPV	2-8	250kHz	Y	O	12,000	105	110
Audio Precision	APx516	2	80kHz	Y	Y	6,000	109	109
PrismSound / Spectral	dScope III	2	200kHz	Y	Y	4,400	105	108
PrismSound / Spectral	dScope M1	2	90kHz	Y	Y	4,400	105	108
Rhode & Schwarz	UPP	2	80kHz	Y	Y	(7000)	100	105
QuantAsylum	QA401	2	80kHz	Y	N	450	101	101

Where are you getting your bandwidth numbers from?

The APX555 analyzer tops out over 1MHz, and its analog gen goes out to over 200kHz.
The QA403 can do 384k (ADC only) for ~180kHz.

 

[welwynnick]

Where are you getting your bandwidth numbers from?
The APX555 analyzer tops out over 1MHz, and its analog gen goes out to over 200kHz.
The QA403 can do 384k (ADC only) for ~180kHz.

Yes the analyser input goes much higher, but I took the DAC output from here:

APx555 B Series

www.ap.com

I think its fair to use the analogue output, so I changed the chart from 80 to 200kHz.
Bandwidth specs weren't easy to come by - often it's just sampling rate - but I doubt it's much of a discriminator for any audio analyser.
In each case, the optimum performance is only achieved over 20 - 20kHz.

 

[KSTR]

Is -6dbFS a good test signal level for mainstream benchmarking ? Or reaching the closer to clipping level is the best practice ?
If a device plays better numbers at -12dBFS is it valid ?

Actually relevant for music recording/playback is the range from -10dB to -30 or -40dBFS. You never run an ADC close to clipping in practice.
Tests at 0dBFS are not very meaningful with regard to real life situation.

However, DACs and ADCs that behave well near 0dBFS typically stay the same or get even better at lower levels. But not always, notably when there is a strong ESS-Hump present...

 

[capslock]

You have to calibrate the Cosmos to actual voltage. 10V nominal was 10.2V for 0 dBFS for me.

As close to clipping without intersample overs is the best.

Maybe I've worked too much with AKM converters, but I have learned to stay away from close to clipping. ESS traditionally do better near fs than AKM devices if you use SINAD or THD+N as your metric. Looking at Amir's IMD vs. input level graph in the first post, I can see the curves flattening. So I suspect that as you get near fs, harmonics will increase but not as quickly as noise decreases.

For optimizing amplifiers, I/V-stages, preamp stages etc. I am always more interested in HD than in noise since I can average all I like. With ESS, though, I am never sure where the sweet spot is because of the hump. Pity it also occurs in ADCiso.

The HPSA (High Performance Signal Analyser) with its outstanding specs is only used for single sine test signals as there is only one notch filter. So the question of the "pure" ADC performance is fully legit.

For multitone the normal ADC path is used, and that path is very comparable in performance to the stripped down signal path in the Cosmos ADC with the ES9822pro, in other words, the Cosmos ADC is on par with AP for anything else than pure sine.

Concur. And as apples to apples goes, we should be comparing the AP in high perf mode to ADCiso with APU, which can be done at 1 and 10 kHz. My understanding is that the AP uses a gliding notch, which is something the APU clearly cannot do.

 

[KSTR]

My understanding is that the AP uses a gliding notch

Yes. The tracking self-adjusting notch is the heart of the whole instrument, besides the analog generator, given their distortion quality.

 

[capslock]

The other point that needs to be understood is an input impedance in the Cosmos mono mode when measuring a component with single-ended RCA output.

View attachment 366618

Though the impedance specified is 1kohm for 2.7V input range, it is valid only in case of balanced input of 1 channel in the stereo mode (R1 + R2 = 1kohm). In the mono mode, L and R inputs go parallel, reducing input balanced impedance to 500 ohm, but this goes again to half, 250 ohm, if we measure any single ended output device, because -In branch (R2, R4) is effectively shorted and we have R1//R3 as a result, thus 250 ohm only. Thus the voltage difference in my previous post. So, be very careful, comparing Cosmos ADC and APxxx distortion test results and level measurements when link level sources have been measured.
------------------------------
I know that Ivan @IVX declares a bit higher value of input impedance in the single-ended mode. I was curious and made measurements of the input impedance. In the single ended mode, 1 channel, I got 673 ohm for the left and 605.5 ohm for the right channel. In the single ended mono mode it makes 318.75 ohm, which is a bit more than 250 ohm estimated above. By simulating with OPA1632, I get 666 ohm in the single ended 1 channel mode and 333 ohm in the mono mode. This is pretty close to the measured values. I am curious if @IVX has anything to add .


SE mode:
View attachment 366701

I am surprised you got something that not only differed from the spec and simulation but was also different between channels. I suppose the input network uses 0.1% resistors just to obtain the CMRR. Even with 1%, I don't see how they could be that different.

 

[restorer-john]

I am surprised you got something that not only differed from the spec and simulation but was also different between channels. I suppose the input network uses 0.1% resistors just to obtain the CMRR. Even with 1%, I don't see how they could be that different.

Unequal source (output R) loading of the two legs to ground in SE/mono.