Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Groff]

Hi,

How low is the noise floor of yours AD converters, regardless of manufacturer specs?

What would be the noise floor we could consider as a pro, low enough for AD stage?

I tested mine with simple test (not shore if the test is good or scientific, but it's a real life situation) I recorded „silence“ with everything unplugged except AD to PC. The result is around -98 dB with some fluctuation up to -96 dB. For 24 bit AD it seems to me a bit higher than I expected. Theoretically that's the noise floor of 16 bit converter.

In the second test I disconnected all my PC hard drives except system disc and all coolers including one for processor (only for 2-3 min) and repeat the test. Results are more or less the same, around -98/96 dB.

I’m really curious about that

Thanks
Best regards

[Patrik T]

I did somthing like you some time ago. I recorded silence from my two avaliable converters: The LavryBlue and the RME Fireface.

Eventhough the RME seemed to have a more flat response throughout the frequency spectrum, the Lavry had lower noisefloor in the "hearable, realistic and crucial region". Both these noisefloors seemed to correspond to what the manufacturers stated in their papers.

I looked at the recording on a Blackmann-presented (If I remember it right) FFT to see the noise floor at random points in the recording. Since I've checked dither amplitudes, noisefloors and looks with this presentation on a FFT, and they seem "right" then, I also trust the FFT when looking at other kinds of noise.

Anyhow - my noisefloor readouts were way below the 96 dB that you state, and they seemed to correspond to the manufacturers data about their gear. I do not know how you checked your recording, but it seems you only talk about the level of noise - perhaps read out on a digital meter in your DAW, and I'm pretty sure that it is not very handy to focus on that value when it comes to the floor of the noise.

[Groff]

Thanks Patrik,

Well ... yes, I did it and looked to the peak meter in Sound Forge 8. I didn't do any further analyzes yet (still in shock with the brand name converter / noise) . I could do it with SForge tools and paste some graph here for discussion. Would that  be valuable?

What's so wrong with DAW (audio editor in my case) peak meter when we looking to the noise floor? There is some inaccuracy on digital meters, I know, but I guess it's a dB or two.

Any advices for testing/analyses?

Thanks again
Regards,
Peter

[Jim Williams]

That test makes about as much sense as evaluating mic preamp noise with nothing plugged into it. What you will hear is the thermal noise generated by the input bias source resistors, which can be as large as 22k ohms.

If you want to "hear" the noise from your A/D
converter, try shorting the inputs, this bypasses
the input resistor's large noise contrubution to
the measurements.

[Patrik T]

Jim Williams wrote on Fri, 31 March 2006 17:37

That test makes about as much sense as evaluating mic preamp noise with nothing plugged into it. What you will hear is the thermal noise generated by the input bias source resistors, which can be as large as 22k ohms. If you want to "hear" the noise from your A/D converter, try shorting the inputs, this bypasses the input resistor's large noise contrubution to the measurements.

Ok, cool. So the plots I get from recording with nothing at the A/D input are actually not relevant. Okidoki.

The difference in the plots have to do with the resitors.

Okidoki.

[danlavry]

Patrik T wrote on Fri, 31 March 2006 19:25

Jim Williams wrote on Fri, 31 March 2006 17:37 That test makes about as much sense as evaluating mic preamp noise with nothing plugged into it. What you will hear is the thermal noise generated by the input bias source resistors, which can be as large as 22k ohms. If you want to "hear" the noise from your A/D converter, try shorting the inputs, this bypasses the input resistor's large noise contrubution to the measurements. Ok, cool. So the plots I get from recording with nothing at the A/D input are actually not relevant. Okidoki. The difference in the plots have to do with the resitors. Okidoki.

The chain is only as strong as the weakest link.
One way to approach audio is to get the ultimate gear for every step in the audio production chain. But such a "blind" approach is very costly, and only marginally better (if any) then the inteligent approach.

The intelligent approach is to learn about the weak links in the chain, and put your money there.

For noise floor bottle knack, look at the microphones and the microphone preamplifier. You will be surprised at what is "waiting for you" there:

Say your micpre spec is 127dBu, which is very good figure! Now, say you use 40dB gain, which means the micpreamp OUTPUT Noise is at 127-40=87dBu. Say your AD has an input range of 24dBu
(pro level as per AES). The dynamic range is 87+24= 111dB.

Now, 111dB is about 18 bits. But that calculation did NOT take into account the microphone noise! Chances are that you are operating at less then 17 bits...

Of course, the less gain setting on the micpre, the more "real" bits, and the more micpre gain, the less real bits.

So if dynamic range is your issue, look first at the micpre. The LavryBlue will not be your limiting factor with 30dB or more gain on the micpre.

For more detail desciption and details look at

  http://www.lavryengineering.com/lavry_forum/viewtopic.php?t= 19

Regards
Dan Lavry
http://www.lavryengineering.com

[Patrik T]

I have the Blue A/D + D/A + the mic pre and I'm very, very happy with everything in that rack.

One AKG C-414 XLS (rather un-noisy microphone) through the blue mic pre to the blue A/D is a remarkably silent operation. And the possible audible noise does not come from the A/D-part, that's for certain.



If that says anything Geoff...

[danlavry]

Groff wrote on Fri, 31 March 2006 00:10

Hi, How low is the noise floor of yours AD converters, regardless of manufacturer specs? Thanks Best regards

I have been thinking about that statement. My first reaction was:
"regardless of manufacturer specs?" I stand behind my specs. my specs are published. What kind of a question is that?

But then I recalled that some people view spec by looking at a device data directly from the IC maker data sheet. Such data reflect the best that the IC maker can do in an "evaluation board environment" with laboratory power supply, with minimal input circuitry... not reflecting of the finished end product...

Such practice of specifying or viewing a product is of course wrong, because the product itself includes additional circuitry which increases the noise, possibly a different supply, layout and so on.

Any self respecting manufacture (me included) specs is about the finished product.

Regards
Dan Lavry
http://www.lavryengineering.com

[Groff]

Gentlemen, thank you very much for input and knowledge you share

Patrik,

Yes, I know, I 'm reading so much about LavryBlue converters, and never saw any complaint. Unluckily, I can’t afford one yet, but I'm planning to do that in the future. Blue is on the highest point on my list of wishes.


Dan,

Thanks for the link and explanation. Of course, I expect more noise from mic and pre, that's obvious, moreover there isn't much music around -96/98 dB anyway. The point is the difference between manufacturer spec and real life, and I'm focusing only to AD stage here. I'm not EE, I'm just user, so I could only imagine how manufacturer do they tests (lab conditions) even for the finished product. I can see only the spec they put on the paper, nothing more or less.

And the spec are:

Dynamic range: 114 dB A weighted (AD + DA)

THD+N: -105 dB (AD), -103 dB (DA)

- no data for S/N ratio
- calibrated at 16 dBFs (both AD & DA)

Now, I did the trick with my PC (mentioned above) to make it silent as possible. Full engaged or just with one system disc, results are pretty close.

Expecting impossible?  

I'll be back ... with some graphs today.

I could put the name of manuf./model easily right now, but I’m trying not to open another can of worms. Not yet.

I'm not native English, so I apologize for clumsiness, but I hope it's clear what's bothering me about AD.

[Groff]

Here's the graph: mono file, 24/44.1, 15 sec.

Patrik was right about DAW peak-lair-meter. More than I expected. Now I feel sooo …    

[Groff]

Long time without replies and comments on graph above.

I would like to hear yours comments on AD noise graph, but I'm not sure if I did the test right and valuable (tool and settings) for analyze. Sorry, mea culpa, I'm newbie in the land of FFT measuring. Can you enlighten me?  

I'm still curious about what I see on my DAW peak meter at - 98/96 dB but not on the FFT graph. How to explain the difference?

Jim mentioned input resistor noise and „shorting the inputs“. What shorting means, and how to do that?


Thanks for patience

Best regards

[danlavry]

Groff wrote on Tue, 11 April 2006 14:52

Long time without replies and comments on graph above. I would like to hear yours comments on AD noise graph, but I'm not sure if I did the test right and valuable (tool and settings) for analyze. Sorry, mea culpa, I'm newbie in the land of FFT measuring. Can you enlighten me?   I'm still curious about what I see on my DAW peak meter at - 98/96 dB but not on the FFT graph. How to explain the difference? Jim mentioned input resistor noise and „shorting the inputs“. What shorting means, and how to do that? Thanks for patience Best regards

I do not know what converter you used to do an FFT, it is certainly not mine. It is not easy to comment without knowing a lot more details. I do not know why you have such higher level up to 200Hz or so.

One has to understand that FFT is a different measurement then dynamic range (and frankly I am not sure what your peak meter does – is it the peak value of an FFT? Peak of dynamic range???)

When you look at an FFT, you get to see the noise PER GIVEN FREQUENCY BAND. Say you have an FFT of 0-22KHz, and you choose 1000 points (called bins). The meaning of it is – your FFT looks at 22000HZ / 1000 points = 22Hz frequency bands. So the left most band is 0-22Hz, the next one is 22-44Hz and so on. The FFT plot shows the noise at each band.

So the 1000 FFT yields detail of 1000 bands, and an 8000 points FFT yields even more detail. But when you specify a single performance number, such as dynamic range, you are, in fact, combining all the energy in all the bins into one number. It is not a simple sum, but the noise of all bins does add up, and the overall number will of course be higher then the single highest bin…

So a dynamic range of, say, 96dB will yield an FFT where the highest point is far bellow 96dB, in fact, for a 44.1KHz case, the FFT is below -139dB. Your FFT shows a dynamic range of far below 96/98dB, but then it is possible that the measurement is “A weighted”. It simply too difficult to say much without a lot of details.

Regarding the shorting of the input resistor: I agree it is a good idea. The resistor noise will be an issue only for real good dynamic range AD, but the additional factor is - an open input circuit has high impedance so it can pick up unwanted signals such as line frequency signal (60HZ).

Shorting means temporary connecting of all the inputs pins to ground with short wires.

Regards
Dan Lavry
http://lavryengineering.com

[Groff]

Many thanks for replay and clarifications Dan.

Quote:

I do not know why you have such higher level up to 200Hz or so.

The power comes first on my mind. I'm using power straight from the outlet (220 here), without any voltage regulator (coming soon). The house is well grounded, but still power can be impure in many ways. Besides I expected some peak around 60 Hz. Again, EVERYTHING in the studio/house, except ADDA converter and PC, was unplugged from mains and wall sockets.

Quote:

(and frankly I am not sure what your peak meter does – is it the peak value of an FFT? Peak of dynamic range???)

It's standard peak meter, measuring highest peak in dB (Sound Forge) not the VU(ppm).

In my case the most problematic area lies in 200 Hz and below. FFT graph is per given frequency band. Clear enough. Now there is another problem. I did some more FFT using the same file, but now with a bin per Hz (S. Forge allows only few fixed FFT sizes, so I set frequency range to match the number of bins).

First graph shows FFT size 512 on 0-512 Hz range and second FFT size 1024 / 0-1024 Hz range. I expected the same values on both graphs per given frequency, but it’s visible that, for example at 200 Hz, the highest value of noise is around -126 dB (FFT 512 graph) and on FFT 1024 graph the value is around -142 dB.

From where comes that difference if the FFT size/Hz proportion is 1:1?






Small step further, to stuck me more. Next two graphs: FFT size 8192 / 8192 Hz range and FFT size 16384 / 16384 Hz range.

Now, my whole problematic phantom area moved from 200 Hz to under 16 Hz and at last under 10 Hz.

Quote:

It simply too difficult to say much without a lot of details.

I wish I could give you more spec and data but all I have found is:

Sample rates: 44.1 - 48k, 88.2 - 96k, 176.4 - 192k (+/-10%);
Frequency response: 10 - 20k (+/- 0.2 dB) at 44.1k
Analog max levels: adjustable between +2 dBu and +26 dBu
Dynamic range: 114 dB A weighted (AD + DA)
THD+N: -105 dB (AD), -103 dB (DA)

There are other things about number of ins/outs and similar but I suppose that’s irrelevant for the problem.



Best regards

[Ronny]

Your total harmonic distortion plus noise spec is -105dB at the AD, your first two graphs show -106dB is the highest noise near zero Hz and your second two show -105dB at the highest level. I fail to see a big discrepancy from your manufacturers specs and what you are determining. BTW, your first graph in this thread is also showing -105dB. What am I missing here, I'm reading that the graphs are showing the specs to be correct, the highest floor is THD+N at -105dB, as stated in the AD manufacturers spec sheet.

[C-J]

Groff wrote on Thu, 13 April 2006 10:11

From where comes that difference if the FFT size/Hz proportion is 1:1?

Groff,

You don't have to match the max freq with the FFT size!
For most accurate results, set your FFT size to 65 536 and your max freq to 300-500 Hz.

C.J., Finland