WiiM Ultra Streamer Preamp Review

[Soniclife]

And here is the important thing. Turning down the volume digitally doesn't increase the noise floor, it just reduces the signal. Just like listening to a quiet part of the music.

There is a presentation from ESS on their volume control that shows lowering the volume (using the chip) also lowers the noise floor, largely maintaining the gap between signal and noise floor.

 

[Miguelón]

And here is why non of that matters.

Dac chips are doing the volume control in either 24 bit or 32 bit DSP. All the bit depth gives you is low noise floor. 16 bit audio has a (digital) noise floor around -96dB. 24 bit around -144dB. 32 bit processing results in a noise floor around -192dB.

In reality though those numbers are not the important ones. Because the analogue electronics in the output stage of even the best measured DACS is not achieving dynamic range of better than about -130dB. Let's say -115dB is typical for a good DAC - eg the Topping D10s at $100. So even the lower spec DACS with a 24bit DSP pipeline - have lower digital noise floor than the analogue noise floor of the output stage.

But even this analogue noise floor is still lower than the level of audibility, assuming your maximum volume level is below that 115dB. In reality you could go 30dB above that, and any DAC noise will still be lower than your room noise floor.

And here is the important thing. Turning down the volume digitally doesn't increase the noise floor, it just reduces the signal. Just like listening to a quiet part of the music.

So If you can't hear the noise at 0dbFS, then you can't hear the noise also at -100dBFS or -150dBFS - the noise is inaudible.


So it doesn't matter that you are "losing resolution" by turning down the volume digitally, because the noise remains inaudible.

But if you reduce the signal digitally you will reduce the bit depth in which the wave would be drawn on analogue domain, or doesn’t?

I mean, with 16 bit/44.1 kHz you can shape a complex signal superposing a lot of frequencies, if you reduce by hypothesis to 8 bits /44.1 kHz you will loose accuracy to shape many overlapping frequencies with different intensities and phases.

16 bit depth is not only due to 96 dB of dynamic range but also to “draw” digitally a continuous function.

Or am I wrong?

On the other hand, viewing that the DAC oversamples to 32 bits there’s no reason to think lowing digital volume will reduce audible information.

My concern was wrong supposing that digital domain only worked at 16 bits depth, and thought only 10 would be disposable when lowering 36 dB

 

[Miguelón]

Really?

What problem is that? Have anything to back. your statement up?

Not including a voltage regulator, at least with one you can choose the way one manage the signal. Digital control can be a choice or not.

All audio interfaces have it, aside by the fact one can choose the volume or equalize on the DAW

 

[antcollinet]

Or am I wrong?

Yes you are wrong.

Higher bit depth gives you less quantisation noise. that is all. This video might help.. I love an excuse for Monty.

 

[Abdo]

As WiiM specifies 32 bits is wide enough to don’t subtract any audible influence even at -50 or -60 dBFS digital volume.

This invalidates my argument, since no meaningful information is lost on the process.

I will repeat tests at home with my girlfriend changing connections and me looking oposite side. Is not very scientific but should be enough to loose references one matching volumes.

In this case only the bad implementation of the RCA at the Ifi DAC, or maybe psychological effect knowing the basic process of attenuating volume digitally.

Thank you for your transparency in your opinions and for acknowledging when you may be wrong or have doubts.

Now I would like to hear your opinion on this new test, whether you would continue to use another DAC or if the Wiim Ultra does the same job.

Thank you.

 

[Miguelón]

Yes you are wrong.

Higher bit depth gives you less quantisation noise. that is all. This video might help.. I love an excuse for Monty.

This has no sense, with two bits you cannot record anything.

A single sinusoid will be only drawn on one single state of intensity, add a second one and it will clip.

Take 10 instruments, each with their own harmonics, which has their own intensity, each instrument also their own intensity at any sample time.

Assuming you can take at least 10 harmonics of each of those instrument, and you can vary the intensity on, for example, 10 different levels for each of the individual components of the Fourier space, you should need at least 1000 levels of signal encoding to reconstruct back the instruments and harmonics from the space. This is around 11 bits.

The main reason you use 16 bits is allowing 65.536 levels of intensity to define you time-intensity space.

If your intensity variable has no accuracy to quantize your intensity function, you have a bad recording. The adequate number was found to be at least, but not less, those 65.000 levels.

You can realize that watching a TV with only 4000 color space, then a 65000 one, and even you’ll notice the difference between the standard 16.000.000 levels.

Same in audio, you need bit depth in the same sense you need bit depth on the RGB space.

 

[Emlin]

This has no sense, with two bits you cannot record anything.

A single sinusoid will be only drawn on one single state of intensity, add a second one and it will clip.

Take 10 instruments, each with their own harmonics, which has their own intensity, each instrument also their own intensity at any sample time.

Assuming you can take at least 10 harmonics of each of those instrument, and you can vary the intensity on, for example, 10 different levels for each of the individual components of the Fourier space, you should need at least 1000 levels of signal encoding to reconstruct back the instruments and harmonics from the space. This is around 11 bits.

The main reason you use 16 bits is allowing 65.536 levels of intensity to define you time-intensity space.

If your intensity variable has no accuracy to quantize your intensity function, you have a bad recording. The adequate number was found to be at least, but not less, those 65.000 levels.

You can realize that watching a TV with only 4000 color space, then a 65000 one, and even you’ll notice the difference between the standard 16.000.000 levels.

Same in audio, you need bit depth in the same sense you need bit depth on the RGB space.

For one or many instruments, there's only one waveform.

 

[Miguelón]

Thank you for your transparency in your opinions and for acknowledging when you may be wrong or have doubts.

Now I would like to hear your opinion on this new test, whether you would continue to use another DAC or if the Wiim Ultra does the same job.

Thank you.

I made the test, I can easily recognize when the digital volume is low in 8 of 10 times, but not yet sure that is digital volume the cause of that.

Is a two variable combination: digital volume and analogue volume. We crossed both to (informally) see if digital at fixed (no EQ so bit perfect) with Ifi Zen preamp at 12-1 o’clock position is audibly different than digital volume at 53 and Ifi Zen preamp at its maximum volume.

The issue is my girlfriend should rearrange the knob at each time she changed, we put a thin sticker on the position that matches better a chosen level at 53 by me (volume of preamp at maximum) when I turned all up the preamp. But the sticker can be not enough accurate and the combination of volumes was not exact…

Even assuming I can hear the difference, the problem is determining which of the two variables (or both) are not transparent.

By one side, it was suggested that volume knobs as attenuators are totally transparent and on the other side we learnt here that digital volume should be also transparent.

So being the test not very accurate, and the two variables totally transparent in practice, the result is inconclusive.

Genelec website, if that helps, advise on the professional monitors help guide, to decrease signal strength on analogue way to match voltage of its monitors, and not to low the digital volume.

As Genelec doesn’t make preamps and no conflict of interest was found on my personal search, I will keep their advice to use the attenuator to match my 96 dB @ 1 Vrms and 104 dB SPL max Genelec monitors.

This will be a combination of the two volumes, the preamp sending 1.2 V rms at -20 dBFS. This let headroom enough to EQ.

 

[antcollinet]

This has no sense, with two bits you cannot record anything.

A single sinusoid will be only drawn on one single state of intensity, add a second one and it will clip.

Take 10 instruments, each with their own harmonics, which has their own intensity, each instrument also their own intensity at any sample time.

Assuming you can take at least 10 harmonics of each of those instrument, and you can vary the intensity on, for example, 10 different levels for each of the individual components of the Fourier space, you should need at least 1000 levels of signal encoding to reconstruct back the instruments and harmonics from the space. This is around 11 bits.

The main reason you use 16 bits is allowing 65.536 levels of intensity to define you time-intensity space.

If your intensity variable has no accuracy to quantize your intensity function, you have a bad recording. The adequate number was found to be at least, but not less, those 65.000 levels.

You can realize that watching a TV with only 4000 color space, then a 65000 one, and even you’ll notice the difference between the standard 16.000.000 levels.

Same in audio, you need bit depth in the same sense you need bit depth on the RGB space.

Did you watch the video? Did you see how a 20kHz sine wave could be perfectly reconstructed with 2 (and a tiny bit) samples per cycle? And of course you can hear the difference with low bit depth (just like you can see with small colour space). But the difference presents itself as noise. As the video I linked demonstrates, when you get down to very low bit rates, you hear a very audible hiss.


Here is another - a recording of a zx spectrum making music with a beeper output that only has two states - on or off. So a single bit. This actually cheats a bit because it is not a fixed sample rate.

It's noisy as hell, but you can hear the music.

 

[Miguelón]

For one or many instruments, there's only one waveform.

The waveform is extremely complex, and your brain decompose in frequency space to reconstruct the instruments as in Fourier coefficients.

You can draw a continuous function, a line, on a reticular page and try to color each square touched by the line, then erase the line.

With the color squares you have a quantized line.

Then you tell someone to do the opposite: draw a line that touch all the squares colored on the page. Then erase the squares.

If you can compare the original and the reconstructed second line, you’ll find a difference. This is due to the so called “quantization error”.

Intuitively you can imagine the thinner the size of the squares, the better you can plot a line and reconstruct them.


On the frequency space of the instruments and its harmonics and intensities that constitute the complex line, this translates on more accurate reproduction of the single components.

 

[Miguelón]

Did you see how a 20kHz sine wave could be perfectly reconstructed with 2 (and a tiny bit) samples per cycle?

I saw the video, but you don’t read my post.

The problem are not single frequencies, one can attribute a 0 to a valley and a 1 to a crest. No problem at all.

Waves are additive, this is the basis of all of the “linear behavior” of audio. If Monty tried to add a second frequency with another intensity it cannot reproduce them.

You will have, for simplicity, 4 different values.

1- Valley of first and valley of second
2- Valley of first and crest of second
3- Crest of first and valley of second
4- Crest of first and crest of second.

Massive simplification but you should multiply any degree of freedom in your human brain Fourier space to have the correct minimum amount of bits.

I ignore the history of the 16 bits, but be sure that 15 is not enough and 17 doesn’t change fidelity.

Knowing how engineers love efficiency, they had came to the best value.

Maybe we’re talking about same concept though, noise is random information and will be noticed not only at sound when instruments are not playing, but also in a difficulty to reconstruct the original signal.

On my side, I still continue to use my external DAC and high levels of digital volume, just because I have the components and subjectively perceived better.

But given the 32 bits of WiiM, the transparency measured by Amir and the fact I couldn’t adequately test my perception with my gear, I retire my recommendation to use external DAC because of lack of good objective arguments .

 

[antcollinet]

The waveform is extremely complex, and your brain decompose in frequency space to reconstruct the instruments as in Fourier coefficients.

You can draw a continuous function, a line, on a reticular page and try to color each square touched by the line, then erase the line.

With the color squares you have a quantized line.

Then you tell someone to do the opposite: draw a line that touch all the squares colored on the page. Then erase the squares.

If you can compare the original and the reconstructed second line, you’ll find a difference. This is due to the so called “quantization error”.

Intuitively you can imagine the thinner the size of the squares, the better you can plot a line and reconstruct them.


On the frequency space of the instruments and its harmonics and intensities that constitute the complex line, this translates on more accurate reproduction of the single components.

You are still thinking stair steps, and ignoring the band limited nature of the signal. Re-watch the Monty video. Then re-watch it again. Keep doing it until you understand what he is saying.

 

[antcollinet]

I saw the video, but you don’t read my post.

The problem are not single frequencies, one can attribute a 0 to a valley and a 1 to a crest. No problem at all.

Waves are additive, this is the basis of all of the “linear behavior” of audio. If Monty tried to add a second frequency with another intensity it cannot reproduce them.

You will have, for simplicity, 4 different values.

1- Valley of first and valley of second
2- Valley of first and crest of second
3- Crest of first and valley of second
4- Crest of first and crest of second.

Massive simplification but you should multiply any degree of freedom in your human brain Fourier space to have the correct minimum amount of bits.

I ignore the history of the 16 bits, but be sure that 15 is not enough and 17 doesn’t change fidelity.

Knowing how engineers love efficiency, they had came to the best value.

Maybe we’re talking about same concept though, noise is random information and will be noticed not only at sound when instruments are not playing, but also in a difficulty to reconstruct the original signal.

On my side, I still continue to use my external DAC and high levels of digital volume, just because I have the components and subjectively perceived better.

But given the 32 bits of WiiM, the transparency measured by Amir and the fact I couldn’t adequately test my perception with my gear, I retire my recommendation to use external DAC because of lack of good objective arguments .

Keep watching. For any band limited waveform represented as samples, there is only one band limited signal that can go through the samples. That is how it works. Note that he shows a square wave. That is already a summation of 10 separate sine waves (for a 20Khz band limit and 1kHz fundamental.

The multitone signal used by Amir - is a massivly complex waveform in the time domain**. But can be perfectly represented (with quantisation noise) and a 16 bit 44.1kHz sample rate. You could perfectly represent it with 8 bits also - just with much higher quantization noise.


(Bear in mind all those other frequencies you are addling must be lower in frequency than the 20kHz.)

**You can see the multitone test signal in both frequency and time domain in this post:

Topping LA90 Discrete Amplifier Review

I would love to buy 6 of them, to be used in my 3 way electronic X overed system. The B200 seems to be a better choice for that use case. The problem is musical, not pure X -Y graphs, how it handles complex signals. There is no problem. The LA90 handles complex signals flawlessly. Here's...

audiosciencereview.com

 

[Miguelón]

Keep watching. For any band limited waveform represented as samples, there is only one band limited signal that can go through the samples. That is how it works. Note that he shows a square wave. That is already a summation of 8 separate sine waves (for a 20Khz band limit and 1kHz fundamental.

The multitone signal used by Amir - is a massivly complex waveform in the time domain**. But can be perfectly represented (with quantisation noise) and a 16 bit 44.1kHz sample rate. You could perfectly represent it with 8 bits also - just with much higher quantization noise.


(Bear in mind all those other frequencies you are addling must be lower in frequency than the 20kHz.)

**You can see the multitone test signal in both frequency and time domain in this post:

Topping LA90 Discrete Amplifier Review

I would love to buy 6 of them, to be used in my 3 way electronic X overed system. The B200 seems to be a better choice for that use case. The problem is musical, not pure X -Y graphs, how it handles complex signals. There is no problem. The LA90 handles complex signals flawlessly. Here's...

audiosciencereview.com

I cannot extract more information from Morty’s video, and I’m not sure that only one band limited signal that can pass thorough the samples. This is an affirmation by the autor, I can search more with time and learn the mathematics.

I always work better with learning the proofs than accepting authority arguments.

I can intuit what you’re telling with the unique possible line, because if one tries to add another different line passing by same samples, adding extra curves will implicate high frequencies out of the band added to the signal.

Thanks for your patience and effort!

As measure as I write, I can figure what’s going on with the key words “band limited”, this creates also a limited subspace of possible representable signals on the time domain.

Even I understand why I didn’t think about that, always studied Fourier in pure mathematics, and there’s no limitation of frequencies: they go from minus to plus infinity (use of negative frequencies is allowed, too).

 

[antcollinet]

This is an affirmation by the autor, I can search more with time and learn the mathematics.

Well that's the trick. If you are not able to trust the demonstration (not just the words) then your only recourse is to get your maths to the poiint where you can follow the maths of shannon-nyquist.

 

[Abdo]

Sorry for the newbie question, maybe this isn't the right place. But after I bought the Wiim Ultra I also changed the speaker cables, it said that I should consider listening after 175 hours, which is when they are in normal condition. Is this true or just another industry hoax? Does softening the cables change the situation?

 

[Doodski]

after I bought the Wiim Ultra I also changed the speaker cables, it said that I should consider listening after 175 hours, which is when they are in normal condition. Is this true or just another industry hoax? Does softening the cables change the situation?

It is a snake-oil hoax. Use quality cables of sufficient gauge for the distance from the amp to the speakers and forget all those urban rumors and snake oil science.

 

[Abdo]

It is a snake-oil hoax. Use quality cables of sufficient gauge for the distance from the amp to the speakers and forget all those urban rumors and snake oil science.

I imagined exactly that... but after I found this forum... my view on these 'audiophiles'... changed completely...

 

[Craigyp]

Good morning, the question I'm always asking but without an answer.
I bought the Wiim Ultra and I love the work it does, so much so that I had a CXN V2 and put the Wiim Ultra as the main one. My question is whether it's worth having a DAC between the Wiim Ultra and my MA 6800 (Mcintosh)?
I saw the notes of the SMSL M500 MKIII. And then I ask, will I have a gain in quality with it compared to what I'm using today?

I have a CXN V2 too and retired that a while back to my office set up. There is a significant improvement using the SMSL M500 MKIII over the built in DAC in the Ultra via USB. I can strongly recommend this and have no hesitation in telling you to go ahead. You won’t regret it. I’d be interested to know how you get on if you decide to go with the SMSL.

 

[Abdo]

I have a CXN V2 too and retired that a while back to my office set up. There is a significant improvement using the SMSL M500 MKIII over the built in DAC in the Ultra via USB. I can strongly recommend this and have no hesitation in telling you to go ahead. You won’t regret it. I’d be interested to know how you get on if you decide to go with the SMSL.

Thanks for your reply, sorry for my English. Do you consider the Wiim Ultra better than CXN correct? But using the SMSL as DAC would be better? ... I wouldn't mind paying 300 for the SMSL I just would like to buy something that I would really get a gain from. I live in Brazil and I'm going to Seattle in 11/20 where my daughter lives and I'll take the opportunity to buy. Thanks.