|
Computer Audio Asylum Music servers and other computer based digital audio technologies. |
|
|
|
Computer Audio Asylum Music servers and other computer based digital audio technologies. |
|
In Reply to: RE: @fmak, clarify your conclusions from src.inifinitewave.ca posted by fmak on August 6, 2007 at 01:47:42:
Infinitewave’s frequency sweep graph for SRC (Best Sinc).
You raise 2 issues:
1. SRC (as in Secret Rabbit Code) results in HF images all over the place. These HF images are beyond aliasing artifacts arising from bandlimit reduction (96k to 44.1k).
2. Quality and measurement of USB based interfaces.
Looking at SRC (Best Sinc) version, I don't see those HF images you’re referring to (see above image - those purple lines are below -97db). (I have by reply post added scale graph depicting colors to db levels). High noise levels are however visible for the low quality SRC algorithms (Linear and ZOH) which must be NOT be used and nor are they recommended! SRC's website suggests they be used ONLY when CPU is a bottleneck. Best Sinc noise levels are below -97db whether this be aliasing artifacts or the HF images you mention.
More facts need to be addressed on src's (as in Sample Rate Converters). This also applies to SRC (note case sensitivity).
a. Algorithms based on real numbers (32 & 64 bit float) are superior to integer only ones. 16 bit allows for 65536 signal variations whilst 24 bit allows for ~16.8m (256 times more). Rounding errors arising from integer algorithms will result in less accurate signal amplitudes. You definitely want to avoid this. SRC, R8Brain, Audition etc use float. Integer multiples of original sampling (as in 44.1 -> 88.2 vs 96) is not a factor ito quality when using correct interpolation - see point c.
b. Upsampling from 44.1k to 96k has no added noise in the original frequency band. That is, noise is not added up to 22.05kHz whilst above this, noise could occur whose extent is determined by design of the src. Because these artifacts are ultrasonic and at very low levels (e.g. for SRC this would be below -97db) they're not a major concern.
c. Whilst much has been said about 'Noise', more needs to said about the 'Signal'. Audition and other srcs don't provide details on interpolation algorithms employed with some using hybrid techniques. SRC uses a technique known as Bandlimited interpolation. This is based on sound mathematical principles and theorems which recreate the analogue waveform from a given set of bandlimited samples. This is most CRITICAL as other techniques are likely to deviate from the analogue waveform resulting in audible interpolation errors. Of course this is a game of trade-offs involving computational cost, signal accuracy and SNR. With SRC, we have superior signal accuracy at acceptable SNR and can be easily computed using latest generation processors. A poorer choice would entail weak signal quality against excellent SNR. Ideal would be a src using Bandlimited interpolation with better SNR but without undue processing overheads.
On USB based devices, this in itself is not a basis to relegate pc based solutions as problematic. Other interfaces are available (PCI, PCIe, Ethernet, Firewire). USB, Ethernet and Firewire are by design bit-perfect. From an audio perspective this is powerful as upsampled data is dispatched to an outboard device or DAC perfectly. This outboard device adds a quality clock... Like any new development, there will be challenges in its implementation but will be superior to SPDIF which is a lossy interface as it offers minimal error correction capabilities.
This post is made possible by the generous support of people like you and our sponsors:
Follow Ups
- RE: @fmak, clarify your conclusions from src.inifinitewave.ca - cics 07:23:18 08/07/07 (1)
- colors to db levels (src.inifinitewave.ca) - cics 07:27:24 08/07/07 (0)
Top of Page ]
[ Contact Us ]
[ Support/Wish List ]
[ Copyright © 2025, Audio Asylum, all rights reserved ]
