LEO BODNAR LBE-1421 Review (Clock Generator)

[TNT]

I thought it would make a difference, but obviously I can't measure that benefit

What measurement did you do to come to this conclusion.

I'd be interested in testing such a word clock

That said, and this is why I created the little table to show the difference between the LEO BODNAR and the SMSL G1, as the later shows 30dB better attenuation at 1Hz. I thought it would make a difference, but obviously I can't measure that benefit. So, at the moment, I guess -70dBc at 1Hz is more than sufficient to get the benefits I measured.

Now, like I said, I'd love testing another DAC with a clock input to check if the same benefits repeat the same way.

Cool. Shouldn't one also look what is going on close to DC. Your charts all ends at 20Hz... or much higher. But then one must realise that it takes a really good measurement system to investigate this or the results are swamped by it. I have a suspicion that nasty things are going on in the sub Hz region...

//

 

[NTTY]

REW stops at 2Hz. No differences with or without the external clock from 2Hz to 20Hz.

 

[VientoB]

Does the clock improve the multitone in the bass region?

 

[Hipster Doofus]

Thanks for your review...

Of course one must have a decent clock / timer for baking cookies for music listening.... mustn't one.

 

[NTTY]

It's not a matter of which tool you're using.
It's just that DAC and ADC clocks have to be sufficiently in sync.
And, of course, that the source signal is built to set each frequency in the middle of a FFT bin.

The AP MT signal Amir shared works perfectly with rectangular window at 48kHz 32k FFT size, and also 192kHz 128k FFT size (which has the exact same bin width).

Of course, if you use a different bin width, it will be a different story.

I suggest you download REW, install and measure yourself with it.

Anyways, I get same results as Amir (eg with the Topping D50III) with the AP MT test tone, and not when using a rectangular window.

Since this discussion started about the noise floor that is higher with Multitone than with a single test tone preventing analysis of potential low level jitter at the foot the tone(s), I guess this is enough to close it.

 

[NTTY]

Does the clock improve the multitone in the bass region?

Nope, see my previous test

 

[NTTY]

That was an interesting test, thank you!

Any chance you could run a multi tone test as well? I wonder if that might show any differences / improvements?

I was reading your test of the SMSL D200. Not unusually, it seemed to get a bit mushy at low frequencies.

View attachment 496862

Replying again with the Multitone test but from a CD Player feeding the D200 via SPDIF Coax. No difference when using the LEO BODNAR (view from 2Hz to 90kHz ) :

The blue trace is the one with the LEO BODNAR. Like I said, the higher noise floor does not allow for the investigation I could do with a single tone.

 

[Nopiano]

Agreed. There's no justification for their pricing.

And presumably no external ‘clock’ will make any meaningful difference to a dCS whether it costs $200 or $20,000?

 

[Laniciffo]

'It is important to note, though, that whatever is the initial pitch error of any CD player, I get the same 3.5ppm when using the LEO BODNAR. That means: if the initial clock precision of the CD player is better than 3.5ppm, then it will be increased to 3.5ppm. '

As you observe that with the Leo Bodnar's ref clock the combined resulting error is a constant 3.5 ppm whatever the source, we can conclude that this error depends only on the DAC and the ADC.
These 2 devices introduce distortion, non-linearities, etc... but those defects can not change the frequency. Only the sampling clocks can.
Therefore 3.5 ppm is the frequency difference between the two reference clocks (ADC vs DAC).

The ADC and DAC clocks are theorelically totally unrelated, yet the measured error is constant. There are 2 options to explain that:
- both frequencies are very stable, but 3.5 ppm apart
- one's frequency error is not constant but the other clock's error perfectly compensates those variations, with a 3.5 ppm offset. Obviously this is impossible as this would require an impossible electrical coupling that would show very obviously in bad SINADs.

Therefore I conclude that the MOTU ans Leo Bodnar ref frequencies are both very stable.
Considering that the MOTU's oscillator probably simply relies on its crystal + its tuning caps to set its frequency while the LB tracks an atomic clock, the MOTU most probably is the one that's very slightly wrong.


Now, the fact that using the 10 MHz ref clock restores an almost perfect pitch means that there is no sample rate converter in the DAC.

The samples simply enter a FIFO in the DAC at the source's pace and leave that buffer at the frequency derived from the 10 MHz ref clock.

However, as commented after your previous article, this WILL cause either an underflow or an overflow: if the water tap provides more water than the sink can evacuate, the bath tub will overflow.
These mismatches necessarily generate either sample repeats (input freq < output freq) or sample skips (input freq > output freq).
In most cases these glitches are several orders of magnitude worse than the small pitch or jitter errors of the sources.
As these are not proportional (% or ppm) errors but discrete sample errors, they may generate huge pitch errors. For example a 11 kHz tone in the signal is normally reproduced with 4 samples when sampled at 44.1 kHz. If you instead output 3 or 5 samples, the frequency error is immense, tens of %, not ppms.

Therefore I still question the whole concept of using a ref clock for a DAC. Using a ref clock is only useful if ALL sampling devices in the chain are synchronized : both the laser pickup AND the DAC.
If they are not, the results will be worse than without the ref clock in almost all cases.

However I may very well be completely wrong !
Actually I would appreciate being proved wrong as otherwise this ref clock input in this Topping DAC kind of ruins their reputation :-(

 

[Rja4000]

Maybe a good idea to check the frequency accuracy of your ADC first ?

Here I fed a Topping E2X2 OTG input with the LBE-1421 output

Note the very high 1.2M FFT size, to maximize frequency accuracy.

This is needed if you want the last digit to be relevant (without using special method).

 

[NTTY]

Sorry, I don't get your test. What was the signal going into your ADC? An analog or digital one? The Topping does not appear to have a clock input, so I'm trying to understand how it relates to what I shared and tested.

Because feeding a super stable PCM digital signal via SPDIF directly to the Motu would get me to 19'997.00Hz (for a requested 19'997.00Hz sine tone), even if I would try to slave the source to the Motu, because there would be no drift big enough to make a difference. In which case, I can use whatever FFT window, by the way...

What would be interesting is that you feed a DAC, that has an input clock, with an unprecise SPDIF signal from a(ny) source, and read the analog output of that DAC with your Topping to see potential frequency deviation when using the Leo Bodnar, or not. Because this is what I did, so I'd be interested.

 

[Rja4000]

Sorry, I don't get your test. What was the signal going into your ADC? An analog or digital one?

The output of the LBE-1421.
It's a square wave, so an analog signal too.

 

[Rja4000]

I'm trying to understand how it relates to what I shared and tested.

You try to measure frequency accuracy, but you don't know the impact of your ADC.

Your ADC may be off by a systematic error, as an example..

If you feed its analog input with a known reference frequency signal -like the LBE-1421 clock square wave- you may check the ADC accuracy.
Then you know better its impact on your DAC frequency measurements.

Well, that's how I would do it, personally.

 

[NTTY]

The output of the LBE-1421.
It's a square wave, so an analog signal too.

Great idea, I'll test that once I find an adaptor!

 

[Rja4000]

adaptor

 

[NTTY]

Yes, I realized yesterday I had some adaptors with my scope:

So I ran the same test as you, but with 19'997Hz since that is the pitch error test frequency I'm used to use. And here we go:

This means a deviation below 1ppm (+0.5ppm, actually). The noise pattern is the same as what you saw, by the way.

But, and there is a big but. I did this test yesterday evening and got caught by my daughter for diner, etc... and I forgot to turn off the computer which therefore continued to perform the same test with the Leo Bodnar feeding the Motu for about 5 hours non-stop. And, when I checked the FFT, I saw a different result:

This is 3.5ppm deviation now. And... surprise, this is exactly what I found when testing the output of the SMSL D200 being fed by the Leo Bodnar. And I remember that day the Motu stayed up for a long time because I did so many tests, with different DACs and CD players, spread throughout the day.

So, again, thanks for suggesting this test. I can now adjust my test protocol and also update the results taking this precise deviation into account. For the future tests, I guess it is easier for me to wait the Motu to heat up - and I'll check how long that means to reach this +3.5ppm deviation - so I can account for it.

 

[Rja4000]

For the future tests, I guess it is easier for me to wait the Motu to heat up - and I'll check how long that means to reach this +3.5ppm deviation - so I can account for it.

Since the Motu has more than 2 inputs, you may also feed the MOTU with the second LBE-1421 output at the same time than your DAC signal, on separate inputs.
So whenever you run this test, you may just compare the LBE clock frequency with your DAC output's measured frequency.
To display both, you may run 2 REW sessions if needed.

 

[NTTY]

Which would indeed allow me to compensate live for the clock deviation, independent of the Motu’s temp.
Nice idea again

 

[Rja4000]

One more comment :
I may be wrong, but I think the FFT bin width is still critical for such high accuracy measurements.
And bin width in Hz is given by [Sampling frequency (Hz)]/[FFT Length]
So beware that when increasing your sampling frequency but keeping the same FFT length you increase bin width and therefore loose accuracy.

I'd stay at 44.1 or 48kHz and 1M or 512k FFT, personally.

 

[NTTY]

One more comment :
I may be wrong, but I think the FFT bin width is still critical for such high accuracy measurements.
And bin width in Hz is given by [Sampling frequency (Hz)]/[FFT Length]
So beware that when increasing your sampling frequency but keeping the same FFT length you increase bin width and therefore loose accuracy.

I'd stay at 44.1 or 48kHz and 1M or 512k FFT, personally.

Yep, I always go for higher FFT Length indeed, when I increase the sampling rate. My standard is 192kHz -> 512kFFT BH7. I tried at 1M and more, and the results were the same.