regal wrote on Fri, 15 September 2006 04:52 |
The good thing about these DAC's is not the non-oversampling its the lack of a filter. Your ears and drivers function as the filter.
These DAC's have become very popular in the DIY community and I thought it interesting that there had been no mention of them at all on this board. I find Lavry's reply informative but he dodged the main benefit. If one were to build the ultimate theoretical DAC, it would have no filter andd the DIY community has basically discovered that a filter isn't needed.
|
A DA without a filter introduces image energy that is nearly as high as the energy in the audible range. In the case of non up-sampling DA, the audio content is centered at the sample rate, twice the sample rate, 3 times…. And it goes up to very high frequencies. The amplitude decay curve is rather slow…
Assuming that you could not hear above 22KHz, you say that the image energy will not be heard. But in fact, the DA output signal, both the audio and the images at frequencies above audio must go through some electronics before reaching your ears. Such electronics may be a preamp, a power amp, a speaker, a headphone amp and a headphone… or in a music production environment the signal may be sent to an analog mixer…
It is one thing to ask an amp, or a speaker (or whatever gear is there) to “process” a signal with energy content that is limited to 20Hz-20KHz (or say 10Hz-50KHz) accurately. It is another thing to expect accurate results when the signal contains relatively high energy at high frequencies. When you try to do that, you will find out that the image energy interferes with the electronics in many way. It degrades the transfer curve, which causes inter modulation at all frequencies including the audible ones, it can ruins circuit badly.
If you did not understand what I said, here is another view, perhaps more intuitive: The DA output signal BEFORE the filter is typically (depending on the signal) made out of a very fast steps, and for some signals, the step amplitude is high (think of say sampling a 10KHz full scale sine wave). Fast changes are made of high frequencies. The electronics after the DA would have a “difficult time” tracking those “fast steps”. An anti imaging filter will “smooth out” the signal so that the signal will move much slower, and without the sudden “jerky” steps. Such a signal happened to represent the original waveform before it ever hit the first microphone…
Then comes the other point: a DA with no up sampling has very non flat amplitude vs. frequency response. In theory, a DA is perfect, because the samples are “zero width”, each sample with proper amplitude. But in practice, zero width samples, or very narrow pulses, carry very little energy, so the outcome will be very weak. A weak signal calls for a lot of amplification, which raises the noise, and that is undesirable.
So instead of narrow pulses, we go for a “stair case” waveform, where each value is held steady until the next sample. That practice (we call it NRZ for “not return to zero). We do so instead of the theoretical narrow pulses (we call them RZ because with a narrow pulse the signal between samples is zero most of the time).
Now, doing NRZ (stair case) solves the noise problem, but it brings on another problem – it causes some attenuation when you get to higher audio frequencies – nearly a dB at 20KHz if I recall (see my paper on Sampling, Over sampling, Aliasing, imaging). That roll off curve (sinX/X shape) is NOT something you can fix with an analog EQ (poles and zeros). When you up sample, that problem goes away. See the graph in the paper I recommended.
I could say much more, but those reasons are more then enough to explain that non up sampling has problems, and why no filter is bad news.
Regards
Dan Lavry
http://www.lavryengineering.com