-24db Precut Sound Quality on Rockboxxed iPod?

[Andhyka]

Hi, I just joined Rockbox forum being Hi-Fi newbie since I got my iPod and decent canalphone (MDR-EX90) one month ago.  

Recently, I installed Rockbox on my iPod after reading some ipod SQ posts on Head-Fi.org, and wow, Crossfeed and HW SQ are awesome! It drains battery and is bit laggy though but it's negligible.

I have a question, does massive pre-cut (-24db) have an impact on sound output? I was thinking of setting my volume to 0db (line out, cleanest quality) and use pre-cut as a volume control to handle my headphone instead. I read somewhere in the forum that input source (44khz, 16 bit) gets upscaled to 32 bit during DSP, and is downscaled back to 16bit for the DAC.

But since the output is compressed (from -24db precut), will it have the same high dynamic range as when it was not cut at all (0db precut, but @ 50-60% volume)?

Edit: I finally settled on this matter. Doing all the above steps is not recommended as it can reduce the clarity/fidelity of audio signal. Either sets the pre-cut level suitable to your headphone, or turn it off entirely.

[pabouk]

Because the pre-cut is applied in the digital domain you lose the dynamic range by lowering the volume using pre-cut. This practically means that the quantization noise will be louder. If I remember it correctly 6 dB corresponds approximately to 1 bit of digital resolution thus by applying -24 dB you would have a digital resolution of 12 bits instead of the original 16 bits.

[Andhyka]

Ouch, 4bit lost in translation is intolerable i presume.  

I thought everything in digital domain will not lose detail and clarity.

[pabouk]

I thought everything in digital domain will not lose detail and clarity.

This could be true only if the DA converter in your player has resolution of at least 16 + 4 = 20 bits and S/N ratio at least 120 dB!

[Andhyka]

Oh well, thanks for the input. I raised this whole issue because the iPod sounds best near/at 0db (according to Rockbox & Head-Fi fellows) and is too loud for headphone, so pre-cutting in equalizer helps but at the cost of reduced/compressed clarity.

The other solution possible is adding amp (which I don't bother spending one cos it is more expensive than the headphone itself) and adding iMod (costs the same as 30GB iPod).

After tweaking little bit, I settled on this:

Volume: -15db
Balance: -3%
Pre-Cut: -9db
Crossfeed (Direct/Cross/Attenuation/Cutoff): 0db, -6db, -6db, 700hz
Hardware EQ (Bass Cutoff/Gain): 105hz/6db - turned on when crossfeed is off

Software EQ is on only for pre-cut, anything else is off.

Ideas/Suggestion/Feedback?

[preglow]

Because the pre-cut is applied in the digital domain you lose the dynamic range by lowering the volume using pre-cut. This practically means that the quantization noise will be louder. If I remember it correctly 6 dB corresponds approximately to 1 bit of digital resolution thus by applying -24 dB you would have a digital resolution of 12 bits instead of the original 16 bits.

This assumes that Rockbox uses 16 bits precision for samples internally, and that is quite simply not true. Most codecs operate with an accuracy at or above 24 bits, so the bits that get shaved away in the precut will probably not be audible in the output anyway.

[mamboman]

thanks for the advise. so i guess it doesn't discard much of the real information from the stream. just invested in an amp and listening through the lineout without the imod, the sound quality is indeed significantly improved but at the expense of portability.

[pabouk]

I do not know much about Rockbox internals so correct me if I am wrong.

Most codecs operate with an accuracy at or above 24 bits, so the bits that get shaved away in the precut will probably not be audible in the output anyway.

I know that Rockbox operates with about 30 bit accuracy at some places but as far as I know the final output to the DAC is always truncated or dithered to the most significant 16 bits.

When you set precut to -24 dB you are approximately "shifting" the original sample value by 4 bits to the less significant bits thus the 4 most significant bits are not used by the original signal anymore. At the final stage before sending the signal to the DAC only the most significant 16 bits are taken (optionally with some information from lower bits by dithering) so in our case only 12 bits from the original signal would be used and the 4 most significant bits will be wasted (zero).

[Andhyka]

Finally, I found the answers what I am looking for!  

http://www.wolfsonmicro.com/products/WM8983
http://www.wolfsonmicro.com/uploads/documents/en/WM8983.pdf

It turns out that Wolfson DAC chip (WM8983) on my iPod Video 80GB uses 24-bit sampling in both its ADC and DAC filters. I am not sure about other iPods/players but this seems to indicate that way. Now I can set pre-cut to its lowest lvl (-24db) and enjoy hi-fi pure sound delivered from my iPod @ 0db volume.  

I'm still looking the answer for this question though: What sampling is used by Rockbox to deliver sound output on its final stage to Wolfson chip? Does Rockbox digitally downsample it from 32-bit back to 16-bit or 24-bit?

[saratoga]

Finally, I found the answers what I am looking for!  

http://www.wolfsonmicro.com/products/WM8983
http://www.wolfsonmicro.com/uploads/documents/en/WM8983.pdf

It turns out that Wolfson DAC chip (WM8983) on my iPod Video 80GB uses 24-bit sampling in both its ADC and DAC filters. I am not sure about other iPods/players but this seems to indicate that way. Now I can set pre-cut to its lowest lvl (-24db) and enjoy hi-fi pure sound delivered from my iPod @ 0db volume.  

That won't work.  You'll still lose 24 dB of SNR.  Anytime you adjust volume in the digital domain, you lose SNR, no exceptions.  Thats why volume is adjusted on the analog side of the DAC.  Don't try and defeat that, you can only make things worse.

I'm still looking the answer for this question though: What sampling is used by Rockbox to deliver sound output on its final stage to Wolfson chip? Does Rockbox digitally downsample it from 32-bit back to 16-bit or 24-bit?

Rockbox uses 44.1KHz as the sampling rate.  I don't know what precision is loaded into the DAC, but I'd imagine its 16 bit.

[Andhyka]

Umm I am unsure of how Rockbox precut works, whether it cuts the bottom half of sound or compress/normalize the whole fidelity. If the latter is what happens, I have to disagree.

Since the EQ/DSP processing done in Rockbox is through 32-bit channel, the pre-cut won't affect the SNR as the audio source is still wholly 16-bit. Yes I know it's already upscaled to 32 bit, but the fact that it is just 16-bit stretched will not affect SNR/fidelity of the original music unless I cut waaaay too much (like minus 96db - 6db * 16-bit).

But the early argument still remains, how does Rockbox cut the music? Does it use compress/normalize algorithm, or simply just cut the lower half of sound fidelity.

Any thoughts?

Hmm, I'm still hoping Rockbox feeds 24-bit audio signal to Wolfson ADC/DAC.

Here's the breakdown:

Digital Music (44khz & 16bit) -> Rockbox DSP (16bit stretched to 32bit) ->
24db Pre-Cut (16bit stretched to 28bit/32bit) -> Final Downsampling (16bit stretched to 21bit/24bit -> Wolfson chip (16bit stretched to 21bit/24bit processed by 24-bit ADC/DAC) -> Pure 16-bit signal audible through headphone

Edit: I revised the breakdown above

[Llorean]

Wouldn't the math be -> 21/24 rather than 20/24, then to 14/16? I'm not really an expert on this, but if a 16 bit signal is scaled to 32bit, you actually use the other 16 bits giving twice the resolution, rather than leaving them empty, don't you? It's for more accurate processing, isn't it?

[Andhyka]

I think 16/24/32 bit and decibels work exponentially, so half the volume of 32-bit is not the same as 16-bit, since 2 ^ 32-bit = 4294967296 and 2 ^ 16-bit = 65536.

But apart from the maths, I can confirm that the 16bit signal stretched to 24bit/32bit and transformed to anything above 16bit itself will not affect its fidelity/SNR. You can check it on Professional audio software like Adobe Audition.

I think let's use an easy example with 4bit/8bit/12bit signal:

1 0 1 0

if we convert this to 12bit signal:

111 000 111 000

The same data is added three times. If it's converted back to 8bit signal:

11 00 11 00

Then downscaled back to 4bit:

1 0 1 0

Its only stretched back and forth, and 16bit signal in 32bit channel is still 16bit (except for X-Fi).  

[Llorean]

Yes, if you make absolutely no changes to the signal, obviously going from 16 bit to a linear multiple (32 bit) and back again will result in identical output. I'm not even sure why you bothered to try to explain that, as it's not relevant once you start changing the signal.

[Andhyka]

Well, that explains the signal pathway I mentioned above. And it's just 16-bit data being transformed over and over until it reaches Wolfson DAC without loss of quality/fidelity, of which I hope Rockbox pre-cut EQ function compresses the whole spectrum instead of cutting the lower fidelity.

Compress vs Pre-cut:

00001100 -> 0010 (data compressed, or in another term, downscaled)

00001100 -> 1100 (the first half of data is deleted or cut, meaning it loses fidelity/SNR)