Last year may be marked in audiophile history as the year CDs started sounding okay after all. For many listeners, the sound of the all-conquering Compact Disc was always removed from the high reproduction quality available from good analogue sources, like reel tape or vinyl LP. Compact Disc sound has gradually improved over almost two decades to a point where it is synonymous with ‘high-fidelity’, even if its limited 16-bit resolution was never going to let it overtake legacy analogue playback.

But a revolutionary, rather than evolutionary, step was witnessed in the world of digits when pro-audio signal processing was applied to CD.

It was discovered that the digital datastream could be ‘upsampled’ to 24-bit/l92kHz for decoding by high-resolution D/A converters, with surprising results. The cosy haven of purist audio was shown that CDs could be made to sound more musical.

Even for the innovator and manufacturer of upsampling hardware, Data Conversion Systems Ltd, the mechanism for the sonic benefits was not easy to explain. As dCS conceded, ‘we have found that upsampling reveals information that is present in the master source, but which is not audible when the CD is played back normally. Upsampling cannot increase the amount of information in a signal and the exact mechanism behind the perceived sonic improvements is currently not entirely clear. We are continuing our research into this subject.’

Following the interest caused by upsampling, questions were raised about not only why it worked, but what constituted upsampling anyway. A comparable process that has been used in digital audio for over a decade is oversampling; and it is confusion with this technique that inspired a return to the subject.

Late last year, Wadia Digital, an authority on high-end digital audio, was frustrated with the audio community’s misunderstanding of the finer points of digital signal processing. ‘Recently’, it stated in a paper for distribution among its dealers, ‘some manufacturers have claimed there that there are large sonic gains to be had by processing CD signals by using sample-rate converters to "up-sample" them to 96kHz or 192kHz sample rates before converting them to analog [sic]. There have been statements made that imply that there is some fundamental advantage to this approach compared with existing CD decoding methods that use over-sampled interpolation filters.

‘When used to convert a CD signal to a higher sample rate, the process of sample rate conversion is mathematically synonymous with over-sampling. Whether this process is performed in a digital filter housed in the same chassis as the D-to-A converter or in a ieparate chassis has little bearing on performance. Any advantage that can be claimed for a rate-conversion system can equally be achieved in a sophisticated over-sampled system such as the Wadia DigiMaster.’

The arrival of the latest revision to Wadia’s top-of-the-range CD player provided us with a good excuse to delve deeper into the minutiae of digital audio nomenclature.

WADIA 860X

Wadia’s revised 860x is a substantial, massive block of metal intent on wringing about as much music out of a CD as possible. From its walled-in floating power transformer to its customizable balanced analogue output, compromise, it can be seen here, is patently not a word in Wadia Digital’s vocabulary. The latest xsuffix upgrade brings 24/96 support to its D-A converter section, accessible as a reference-grade unit in its own right through a selection of pro-grade interfaces on the rear of the dreadnought chassis. This allows the 860x to be used as D-A converter for other sources like DVD, DAB radio or MPEG/ATRAC audio. Likewise, the transport section can be used alone; like the DAC, this hosts AES/EBU balanced digital, BNC and RCA coaxial S/PDIF, and ST glass-optical connectors. Even if no other digital sources want for the Wadia D-A treatment, a digital signal processor can be inserted into the loop, taking digits out of the 860x for processing, before returning back home for conversion into analogue. Room-correction equalization is one use that springs to mind. One could even insert a digital-to-digital converter here, that piece of hardware that is also known as an upsampler...

Differences in terminology between upsampling and oversampling may in part be down to semantics, but word-play doesn’t explain differences in perceived sound from ‘mathematically identical’ processes.

To compare the results of the not-so-very-different techniques of oversampling and upsampling, I spent some time listening to the Wadia 860x CD player, and to the dCS Purcell/Delius D-D and D-A converters [HFN/RR Dec ‘99 and April 2000].

Wadia is proud of the high quality of its oversampling digital interpolation filters, trademarked under the name DigiMaster. This uses dual Motorola DSP chips configured with a proprietary algorithm, and engineered, as Wadia puts it, for music. Its Resolution Enhancement Technology adds uncorrelated dither to the extracted digital data, a random noise signal added to linearize the least significant bit (LSB) of the datastream, while expanding the 16-bit data to 24-bit. The amplitude distribution used is Triangular Probability Density Function (TPDF).

The Wadia was used as CD transport in both configurations; both Wadia 860x and dCS Delius were balanced-connected to the preamp. A good ‘window’ into the front-ends’ performance came from Chord Electronics CPA 3200/SBM 1200C pre- and power amplifiers, and Quad 989 and Harbeth LS3/5A loudspeakers. Speaker cable was Ecosse Reference Ultima, while interconnects included vdH The Second balanced, and Chord Co Anthem and Chorus balanced.

In listening tests the Wadia 860x held itself up as a first-class player of Compact Discs one of the finest available on the market today. Just focusing on bass quality for a moment, it showed an extremely tight yet fluid quality with fine texturing that had a good, tangible ‘organic’ feel to it. It carried great weight. Overall presentation was forward of neutral, in the sense that the soundstage started in-line with or even in front of the axis of the speakers. I liked its midband textures, too, which gave corporeal body to vocalists and solidity to instruments like cello and violin. High frequency extension was smooth and refined, and rarely troublesome.

Switching to the dCS upsampled version of events, the soundstage fell back several feet relative to Wadia, now describing an arc that naturally filled one side of the room. In fact the effect was comparable to the entire wall being removed, and musicians now performing live from just outside the building. The colourful richness of the Wadia had disappeared, and with it the vinyl-like rhythmic underpinning that made music so engaging. In its place I found clean, pure sound, bereft of the sharpness and hemmed-in quality of CD playback.

For me, the biggest difference between the Wadia and the dCS sound was in the upper frequency extension. While the 860x played CDs with beautiful aplomb, they still sounded like CDs — CD played with consummate excellence, but nevertheless still a CD — where high playback volumes could become less relaxing in long spells. With Purcell and Delius on form, the effect was less like listening to a CD playing at all. Harmonics seemed to extend beyond the upper threshold with ease, in way I only remembered before from native 24/96, SACD, and analogue sources. Subtle-yet-insidious high-frequency distortion was reduced enough to improve extended, relaxed, listening. Interestingly, I found the ‘sense’ of music better unravelled, as if a recording had better phase-alignment throughout its bandwidth.

This was not intended to be a shoot-out between two company’s technology, nor even between their respective products. For the moment, I find myself perplexed again that high-powered digital signal processing in the upsampling instance can make Compact Disc sound so palatable, allowing it to lose some of its hallmarked ‘closed-in’ signature and gain the air and ease that makes music so enjoyable.

Several other companies are now offering units billed as upsamplers, either built within the D-A converter, or as separate add-on units. Perpetual Technologies, MSB Technology and Dodson Audio, all in the US, are advertising such technology, and we hope to investigate this trend to discover if the technique does universally improve CD playback or if it is confined to particular applications of the technique.

While there is certainly overlap (but never ‘uplap’!) in the use of the terms ‘oversampling’ and ‘upsampling’, some guidelines can be given to differentiate the processes.

Oversampling is typically used to describe a technique used when transferring beiween the analogue and digital domain, where a signal is sampled many times over and above that actually required by the sampling frequency.

Oversampling in the context of the D-A process involves multiplying the sampling frequency by a whole number, typically between 4 and 32, or even higher. For example, in ‘8x oversampling’, CD’s base rate of 4.4.1kHz is raised to 352.8kHz by introducing seven new ‘empty’ samples between the original data samples. These new samples, though, are often not just empty strings of noughts, but based on mathematical models to assist the DAC to work more linearly with the extracted data.

Oversampling, as well as easing the workload of the anti-aliasing filter, which can now operate more gently at a higher frequency, can also reduce distortion created when those analogue signals are first turned from continuous, analogue waveforms into stepped, digital, stair-like curves. This quantization noise is now spread over a larger band after oversampling, and can even be somewhat shifted out of the audible envelope by the technique of noise-shaping. Sony/Philips’ Direct Stream Digital, as used

in SACD, takes this idea to its limit, in order to dump high levels of digital noise up to higher frequencies than are not directly audible.

Upsampling is a solely digital domain process where the data stream is also stretched out by interpolation — guessing the points in between, again mathematically — and is typically used to refer to small, non-integer changes, such as from 44.1 kHz to 48kHz. When the change is larger than this, such as 44.1 kHz to 192kHz, ‘upsampling’ is a more popular term.

'There is apparently no extra information in the upsampled signal that was not present in the initial signal,’ says Mike Story of dCS. ‘With a 44.1 kS/s input, both the input data stream and the upsampled data stream will only contain a spectrum that must be between 0 and 22.05 kHz and is probably only between 0 and 20kHz.'

'This conventional analysis starts from the viewpoint that the behaviour of the ear can be described in mathematical terms using Fourier analysis. This assumption is probably pretty good — it means we are interested in frequency responses, for example, and these do provide good guides to the performance of equipment and to descriptions of what we hear. The analysis was right at the heart of the definition of the audio coding used on CDs.'

‘For those working with audio, it is also apparent that thearies based on these descriptions are not completely adequate, and that there can be significant differences in the performances of pieces of equipment with similar "conventional" specifications. It seems that two things are going on here: the ear may have more than one mechanism at work; and sine waves may not be the best function to use as the basis for analysis. On the mechanism front, it seems highly likely that the ear has a sound localisation mechanism ("where is it?") that is fast, and independent of the mechanism that says "it’s a violin", and that is related to transient response. There may also be a third mechanism at work. On the analysis front, it may be that some form of wavelet is the best basis for mathematical modelling. The problem here is that sine-wave theory is relatively simple, and has been fully worked out by generations of mathematicians, following on from Fourier. Wavelet maths is just plain hard work, and does not yet have anything like such a solid core of mathematical results to call upon. Our ears, however, are not waiting.’