That's language for you! That's life for you! It happens everywhere and in all languages.. it's how humans work. It also seems to cause more problems for those who think in only technical terms and don't like vagueness. That's a kind of personality trait - not bad or good, just a trait.
DIY audio is open to all and also incorporates those mainly interested in music, dabbling in the DIY and picking up vague notions allong the way.
I'd say that's a bit unfair, albeit he's creating a case where one's individual lexicon is correct, which ultimately means no one knows what another person's talking about. Defeats the point of a language!
I argued it's fine if people don't have a perfect definition of oversampling versus upsampling because those have a lot of overlap, but if one wants to call a sigma delta converter non-oversampling, then one is incorrect.
That depends on the reason for rejecting it. If it has something to do with disliking steep filters or not being satisfied with the performance of the oversampling filters used in DAC chips, then there need not be any objection against using a sigma-delta modulator with unity signal transfer function and huge oversampling factor.
Typical weaknesses of FIR interpolation filters include:
1. Hard clipping on peak sample normalized recordings (intersample overshoots); this whole issue wouldn't exist if digital recordings were mastered a bit softer, but many of them are not.
2. Non-linear behaviour and poor filtering of very-low-level signals due to rounding of intermediate products.
3. Pre-echoes in linear-phase filters with non-zero passband ripple.
4. Transition bands up to 0.55 fs, allowing imaging between 0.5 fs and 0.55 fs.
You can avoid issues 1, 2 and 4 and reduce 3 to a negligible level by putting your own filter in an FPGA, like 3lite has done.
Perhaps I was a bit harsh.
It's not just the sigma-delta converter though, the OP referenced a DAC with both sample rate conversion to 384 kHz and a sigma-delta converter, and that started the whole discussion. Pretty hard to make a reasonable argument that a box is non-oversampling when it oversamples in two different places
.Do you think #1 isn't accounted for in most implementations? Another user shared some data at full-scale, (can't find the thread) and it looked like the AKM DACs didn't clip, but maybe the ESS part did.
According to Benchmark it isn't:
Intersample Overs in CD Recordings - Benchmark Media Systems, Inc.
"Faulty D/A and SRC Chips
Every D/A chip and SRC chip that we have tested here at Benchmark has an intersample clipping problem! To the best of our knowledge, no chip manufacturer has adequately addressed this problem. For this reason, virtually every audio device on the market has an intersample overload problem. This problem is most noticeable when playing 44.1 kHz sample rates."
By the way, the main problem is the fact that peak sample normalized recordings are slightly louder than full scale. When the loudest samples are at full scale, the interpolated part in between the samples can be a bit higher.
Intersample Overs in CD Recordings - Benchmark Media Systems, Inc.
"Faulty D/A and SRC Chips
Every D/A chip and SRC chip that we have tested here at Benchmark has an intersample clipping problem! To the best of our knowledge, no chip manufacturer has adequately addressed this problem. For this reason, virtually every audio device on the market has an intersample overload problem. This problem is most noticeable when playing 44.1 kHz sample rates."
By the way, the main problem is the fact that peak sample normalized recordings are slightly louder than full scale. When the loudest samples are at full scale, the interpolated part in between the samples can be a bit higher.
Last edited:
I see. They talk about their high headroom filter but they are using a TI SRC4392 in front of everything, so their real solution is just to attenuate by 3dB. I wonder if the attenuator in the SRC4392 is before or after the ASRC.
It's not just the sigma-delta converter though, the OP referenced a DAC with both sample rate conversion to 384 kHz and a sigma-delta converter, and that started the whole discussion. Pretty hard to make a reasonable argument that a box is non-oversampling when it oversamples in two different places
Can't/won't argue that one.
Yeah, they have a Spartan 6 I think in there. That makes sense if you have a separate SPDIF receiver.
They probably attenuate the USB input in the XMOS receiver, but hard to say with the SPDIF input. SPDIF RX is done by SRC4392, so it seems like a pain to send the data out via I2S to the FPGA and then back in to convert it, but maybe that is just what they do.
As far as I know you can't do that (insufficient number of I2S interfaces on the SRC4392 to get a signal out of it, process it, put it back in and read it out again). I ended up using a separate DIX4192 in my valve DAC. No idea how Benchmark solved it without a separate SPDIF/AES3 interface.
While most intersample overs are just the result of clipping on the original file, I think it's possible to have sequences that are below 0 dBFS and don't violate Nyquist, but reconstruct above that. Very unlikely in the real world out of an ADC, but probably possible from post-processed files. I'm not sure what criteria Benchmark uses to differentiate them from garden-variety clipping.
With peak sample normalization the recording as such is not clipped, it just gets clipped when the reconstruction filter has no headroom. When the recording itself is already clipped, you obviously can't solve that, although you can make it even worse by passing it through an interpolation filter with no headroom. It's a pity, but many recordings are peak sample normalized to 0 dBFS (and many others are already clipped by themselves, which is even worse of course).
Why do you prefer DACs to be designed such that they can't handle normal music recordings properly?
Last edited:
Did you try this one? No op amp for PCM5102 and I'd expect it sounds far better than the price suggests. SA9227+PCM5102A 32BIT/384KHZ USB DAC/HIFI Asynchronous Decoder WIN7/WIN8 | eBay
With Foobar and the Resampler-V component, you can experience what different upsampling filters sound like using this DAC.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.