Sound Quality difference between inputs

Interesting.  What noise levels are involved, compared with the sensitivity of the instrumentation?

That depended on the noise source and configuration of the instrument. In one case the noise was about 10x the signal strength, but transitory and only affecting 3-4μs.  In a more common scenario with the equipment it was typically 0.25% to 5% of signal, but in this same scenario we would also have between 0.1% to 25% of electronic noise from other sources.  There were other scenarios and configurations where it was possible to reduce the noise coupling to irrelevant levels.

In another case with a particular external power supply in use, if the user configured the instrument's cable layout wrongly it could be more than 100x the level of the signal.  This was a costly mistake as the coupling of the noise to the analogue input of the ADC was strong enough to kill the ADC despite fast protection diodes and double screened cables! (£2240 down the drain in the early 90's).

Hmm - I'd be surprised if most people get that level of noise in their homes.  I WHT 5% shouldn't be a problem, though.

Beachcomber posted:

I would have thought that NRZ (and even NRZI) and Manchester are mutually exclusive...  

I would not say they are  mutually exclusive - but I think I see what you mean - NRZ is value and Manchester encoding is transition and time state - I always think of  Manchester encoding as an evolution of NRZ or an application of NRZ . NRZ is a method of representing over an analogue transmission line or equivalent a digital value by assigning a digital value to a voltage such as a binary '1' to a +ve voltage and and binary '0' to a -ve voltage.  But NRZ in itself is not that reliable for higher speed communications and requires additional effort if the values represented are a synchronous bit stream - hence Manchester encoding.

Manchester encoding builds on top of  NRZ by creating a new encoding method (and breaks the absolute assertion of a physical voltage digital mapping at a point in time) by looking at transitions rather than the pure values of NRZ  to create  a synchronous self clocking bit stream, where the clock and the encoded/mapped digital values is extracted from the NRZ transition changes and the previous clocked state rather than the absolute voltage values. This makes it easier for higher speed serial data to be communicated within synchronisation.

But  that is my point the so called digital 'value' is derived from various voltages - its an analogue world at the level we are talking about.... When we are dealing with pulses and transitions of high speed analogue pulses - energy is required and is consumed for conveying these transitions. This energy can couple into other system functions - and I say this is why we hear differences - but it itself has no direct bearing on the digital values communicated - other than they need to be present.

Agree about the design of equipment to handle this - and TI have in my opinion a  nice engineering white paper on managing and reducing noise from this switching energy from USB and Ethernet - I have posted the link to it a few times on the forum - but Google can search it out.

Simon

Beachcomber posted:

Hmm - I'd be surprised if most people get that level of noise in their homes.  I WHT 5% shouldn't be a problem, though.

I'm pretty sure 5% would degrade an audio signal more than MP3 encoding!

I believe in the current generation of digital sources, breakthrough of digital noise (clocks, processing load etc.) into the analogue circuitry could easily be of the order of 0.1% peak to peak (as spikes, and much lower as RMS) before the analogue filtration, even with high quality audio component designs.  The electronic perturbations of the analogue electronics by even this amount is likely give subtle but audible changes to the perceived sound.

Don't forget that the digital switching noise fingerprint of different sources will vary, as will the amplitude of the modulation (particularly in the current domain).  Even USB sticks can have very different internal data layouts due to the organisation of their memory chips, and this will have a profound effect on the pattern noise from their internal clocks.

OTOH I would have thought that it would be possible to remove that 5% rather more easily than a higher value.

With NRZ the voltage does not return to zero at each bit, so you could have a series of 1s and the voltage never (in theory) changes.  You can't get a reliable clock rate from that.  With Manchester encoding the voltage does return to 0, on every bit, giving you a clock.  NRZI does add a wrinkle, in that a voltage change indicates a 1, and no change indicates a 0.  So a series of 1s is easy to detect, but a string of 0s is still hard.  Manchester involves a signal change for every bit, and direction of the change determines whether it is a 1 or 0.  Which means that you have a clock embedded in the signal, with the disadvantage that you have a higher frequency of change.