[jefemeister]

So if I understand, your plots are in this order:

1. Toshiba, 16 bit
2. Reference File, 16 bit
3. Toshiba, 24 bit
4. Reference File, 24 bit

If this is the case, the 16 bit Toshiba data actually looks pretty good except for the last "hump." You can also see some differences in the curvatures between points. These files are not the same.

I'm confused by the 24 bit part. How did you get a 24 bit reference file? It doesn't appear to be the same as the 16 bit one at all. I'm assuming this is a different section of the signal entirely? It also kind of looks like the sample rate is 2x.

The recordings can not be longer than the reference file. You have control over the length of recording. you should probably start recording while the track is playing, then press track backwards, and then stop recording after the track finishes playing. you should be able to grab the section inbetween the silence. You may have to offset the data a couple samples for some reason, but I didn't find this necessary.

Does your recording from the Toshiba match mine?

 

[Mr.Radar]

Quote:

Originally Posted by jefemeister So if I understand, your plots are in this order: 1. Toshiba, 16 bit 2. Reference File, 16 bit 3. Toshiba, 24 bit 4. Reference File, 24 bit

Nope. The reference file is always 16-bit.

Quote:

I'm confused by the 24 bit part. How did you get a 24 bit reference file? It doesn't appear to be the same as the 16 bit one at all. I'm assuming this is a different section of the signal entirely? It also kind of looks like the sample rate is 2x.

I just randomly picked spots (not the most scientific, I know) and I accidentally changed the zoom level on the second pic.

Quote:

The recordings can not be longer than the reference file. You have control over the length of recording.

I said the reference was always longer than the recording, by about a second.

Quote:

Does your recording from the Toshiba match mine?

I'll check when I come back (I have to go to my brother's band concert).

 

[jefemeister]

Quote:

Originally Posted by Mr.Radar I said the reference was always longer than the recording, by about a second.

That's seriously weird. Probably why I read it backwards. . .

 

[Wodgy]

Just FYI to possibly clarify why RMAA doesn't detect any of these errors, there's an interesting thread in the portable audio forum where an ex-Creek engineer measures the iPod using RMAA. This is interesting because there are full Stereophile measurements for the iPod available made using real Audio Precision gear, and RMAA manages to miss a lot of the details those measurements detect.

RMAA seems to be pretty inaccurate in general, though it is useful for getting a general impression of what's up.

 

[antonik]

Quote:

Originally Posted by Wodgy Just FYI to possibly clarify why RMAA doesn't detect any of these errors, there's an interesting thread in the portable audio forum where an ex-Creek engineer measures the iPod using RMAA. This is interesting because there are full Stereophile measurements for the iPod available made using real Audio Precision gear, and RMAA manages to miss a lot of the details those measurements detect. RMAA seems to be pretty inaccurate in general, though it is useful for getting a general impression of what's up.

There is nothing inherently "inaccurate" about RMAA, and my data in another thread rather confirm it. There are limitations you need to be aware of (as with any measuring tool) . If you use a tool incorrectly you'll get incorrect results, as simple as this. I am using different versions of RMAA for about 3 years now and often "in parallel" with some other "proper" measuring gear, so I've got a pretty good idea what to expect from it.

RMAA results in the analogue domain are limited mostly by the quality of the card used and there are some other limitations - for instance it is not very useful for comparing digital output data (unless something seriously wrong with a really bad digital processing going somewhere) as it uses a dithered signal and any well executed resampling could go unnoticed. For tests of a digital output the best "free" tool is actually EAC - it can easily prove if the datastreem is not bit-perfect.

Alex

 

[jefemeister]

Quote:

Originally Posted by antonik For tests of a digital output the best "free" tool is actually EAC - it can easily prove if the datastreem is not bit-perfect.

That's basically what this thread is about. If you get a chance, please read through all of this and let us know what you think. I'm getting one set of results for the Toshiba's digital out using EAC and files recorded with my soundcard's digital input. Mr.Radar is getting a completely different result using RMAA.

 

[antonik]

Quote:

Originally Posted by jefemeister That's basically what this thread is about. If you get a chance, please read through all of this and let us know what you think. I'm getting one set of results for the Toshiba's digital out using EAC and files recorded with my soundcard's digital input. Mr.Radar is getting a completely different result using RMAA.

I've tried to follow the thread but it proved to be a bit difficult

.

If the S/PDIF output of Toshiba is bit-perfect, than there is no need for further questioning what's happening where. If it is NOT bit perfect i.e there is a difference in the data recorded from the digital output compared to the data recovered from the same disk by EAC, it means that the data gets digitally processed by DSP in Toshiba for whatever reason. As far as I understand the output data still 44.1 kHz 16 bit? Not 24 bit by chance? S/PDIF would happily accomodate either. Usually the digital output should be taken before any processing done, however it is not always the case, especially with DVD players. On the other hand, modern DSP's are very good at producing quite perfect measurements on simple signals. If somebody can have a look at the digital output of Toshiba with Digicheck from RMA it will show the number of bits of actual data. Other way is to record the output in 24 bit format and check if 8 lower bits are zeros. If not, than it is definitely processed and filtered.

I do believe that any large imperfections of digital filtering frequency response would show on RMAA measurement, however usually modern digital filters would have very small deviations - much less than 0.1 dB and this would be not visible on RMAA graph. Use of compensation in the analogue stages for frequency deviations in the digital part is very unlikely. It is possible and was used occasionaly in the past, however in analogue domain you can only make reasonably rough correction and RMAA graphs are far too perfectly looking for that.

Alex

 

[jefemeister]

I am measuring significant differences between the digital out and EAC. I get the exact same output from the Tohiba every time even though it is clearly not bit perfect. I do, however, get bit perfect (also repeatable) results from a Pioneer DVD player. Data from the Toshiba is definitely 44.1/16. My best guess is that this the work of an ASRC, but the output is shown to be so crappy that the reults almost defy belief. In fact, without some major signal corrections in the analog stage, this player would have the worst freq-response I have ever seen. It therefore would absolutely NOT be a good choice for use with an external DAC. Again, someone (Mr.Radar seems the likely choice) needs to verify my data is correct and I don't just have a bad unit.

Quote:

Originally Posted by jefemeister ... I recorded 30 seconds of whitenoise and plotted FFT(Toshiba)/FFT(EAC) to give a pretty clear indication of the Toshiba's effect. It appears that this filtering is not FIR as seen by the non-linear phase. -6dB point of magnitude is approximately 6.8kHz.

 

[antonik]

Quote:

Originally Posted by jefemeister Again, someone (Mr.Radar seems the likely choice) needs to verify my data is correct and I don't just have a bad unit.

There is always a possibility that there are different versions of firmware responsible for the DSP algorithms are involved - in this case two different players could give very different results. Did you try to measure yours with RMAA?

Alex

 

[jefemeister]

Quote:

Originally Posted by antonik Did you try to measure yours with RMAA?

I mean to, but I have to look at it in more detail to figure it out. The biggest thing I'm sruggling with is that I don't think you can measure freq response effectively this way. Audio Precision, etc does it by playing single-freq sines and recording the output. If you play many sines at once, you get all kinds of inter-harmonics and crap to deal with. Maybe RMAA has a solution (like wait states and timeouts) but, like I said, I don't currently know much about it.

 

[jefemeister]

Actually, now that I think more about that plot I quoted above, could be linear phase meaning that the filtering is probably FIR based. The curvature around the min/max occurs in the transition region of the magnitude. This is expected for FIRs. The "straight" should still be straighter (

) but the curvature can probably be attributed to the whitenoise method of measurement. Doesn't really matter too much as we're mostly concerned with the magnitude but I don't want to be responsible for any misinformation.

 

[Glassman]

I got some new measurements of my Pioneer's S/PDIF output.. this time using 24/48 LPCM track on some DVD-V.. very interesting results, there is the same -0.034dB attenuation as when playing redbook CDs, but the 24bit data are truncated to 16bit before the attenuation! so you basicaly get 16bit data in 24bit S/PDIF stream shifted by ~5 LSB.. however this likely indicates that the attenuation is bounded just with the S/PDIF output and doesn't affect I2S streams inside the player.. I'll check that next week..

 

[Duncan]

Quote:

Originally Posted by jefemeister ...Basically there are two kinds of digital filter topologies: FIR (Finite-Impulse Response) and IIR (Infinite-Impulse Response). IIR's use feedback in their equations (each output is based off both previous inputs AND outputs) while FIR's do not (each output is based off previous inputs only). This means that IIR filters can be shorter and thus faster to implement, but will distort the phase of the signal in the process. FIR filters, by definition, are linear-phase which means the only phase distortion is a simple time-delay over the entire signal which is trivial. But they are longer than IIR's and require more time to compute.

You seem very in the know jefemeister...

Any idea what an 'Z-Filter' is then? (I reportedly have one on this aged Marantz player, so i'm not expecting to be told that it is the new messiah of filtering or anything, just curious as to which of those philiosophies it follows... i've googled (including groups) and found nothing)

Thanks

 

[dwk]

I've just noticed this thread, and have only scanned the posts, so this might have been touched on before.

Could the problem be that the Toshiba is not handling the pre-emphasis spec properly? The order of magnitude from what I can find on the web is 9dB at 15kHz, which matches jefemeister's latest plot. It looks like the Toshiba might be applying digital de-emphasis to material when it shouldn't. This would certainly be a bizarre thing to do, since virtually no cd's are recorded with emphasis, but it would at least be a 'somewhat sane' explanation.

I haven't yet found a real description/plot of the de-/emphasis curves, though, so I could easily be off-base.

 

[jefemeister]

Quote:

Originally Posted by dwk It looks like the Toshiba might be applying digital de-emphasis to material when it shouldn't. This would certainly be a bizarre thing to do, since virtually no cd's are recorded with emphasis, but it would at least be a 'somewhat sane' explanation.

I was thinking the same thing, so i went back and checked that none of the CDs had emphasis (EAC will tell you when you load a disc). I figured that always being locked into de-emphasis would not be a possibility, but I'm open to anything at this point. My copy of Pink Floyd's The Wall has pre-emphasis. I will try recording a section of that and see what happens.

Mr.Radar, we're still waiting on a comparison of your recordings to mine when you get a chance.

Duncan,
I have to take off right now, but I will try to answer your Z-filter question tomorrow.