dbx Recording Technology Series
Noise reduction explained.
Dbx has been around for a good few years now, and, in its time, has attracted its fair share of brickbats from dedicated audiophiles and Hi-Fi journals. There's a paradox in this, for dbx has established itself as one of the rock-steady favourites of professional recording studios, alongside or in place of Dolby A.
However, faced with the public's insatiable appetite for greater degrees of noise reduction in consumer cassette decks, and the rapid growth of the 'alternative' noise reduction industry, I suspect that dbx were practically given an ultimatum to improve on their act, or, at least, make their products more acceptable to well-tuned ears. Their Recording Technology Series of noise reduction units represents one solution to these requirements, and a press reception at the Savoy Hotel in June saw the first British airing of the refurbished dbx.
There's much mystique attached to noise reduction systems, when really all they're doing is coding the signal in a form that on decoding will resurrect the music exactly as it was originally. The coding can be as simple as 2:1 compression, so that a 12 dB increase in dynamic level is reduced to just 6dB, or as complex as A/D conversion, so that the signal is actually subjected to the ultimate form of coding, i.e., transformed into binary code. Whatever coding is used, tape hiss and machine noise will be pushed down in dynamic level when the off-tape signal is decoded.
In the case of compansion systems using 1:2 expansion, the noise generated during the recording stage will also be expanded, but, since it wasn't part of the coded signal (hopefully), the noise components will be pushed below the limits of audibility. With digital decoding, i.e., D/A conversion, only the binary code corresponding to the signal is transformed back to its original analogue state, and, as a consequence, noise just gets ignored. Neither coding/de-coding technique is exactly easy to implement. With digital techniques it's necessary to use a sampling rate that's at least double the top end of the frequency response and use at least 12-bit conversion to reduce quantisation error. However, once the digital modus operandi has been worked out, there aren't many things that can go wrong with the operation of the system, as the heart of the process - A/D and D/A conversion - goes on transforming information regardless of the dynamic or spectral content of the music. With compansion systems, this is where the problems really start.
Dolby B has enabled good cassette decks to achieve signal-to-noise ratios in excess of 60 dB, a figure totally beyond the expectations of Philips when they first introduced the compact cassette in the late 1960s. However, Dolby B's 10 dB of tape hiss reduction is a compromise between what can be achieved with compansion systems (30 to 35 dB improvement of S/N ratio) and the problems encountered in the use of such comparatively extreme signal processing. It is generally conceded that the more compression/expansion there is applied to an audio signal, the more likely it is that modulation of the noise accompanying the signal by the compander will become audible. It's this modulation that produces the notorious 'breathing' or 'pumping' often considered as being the biggest drawback of compansion systems. However, there are a number of ways of getting around this.
The complex expedient adopted by Dolby in their professional system, Dolby A, was to split the audio spectrum into four separate bands, less than 80 Hz, 80 Hz to 3 kHz, 3 kHz to 9 kHz, and greater than 9 kHz, derive control voltages from each band, and then compress each band separately. During playback, the signal is again divided into bands and each is expanded back to its original level. The end product is 10 to 12 dB of noise reduction, which is fine for recording studios using professional tape equipment already capable of S/N ratios of 75 dB or thereabouts, but hardly a 'quantity' improvement on the noise reduction offered by Dolby B, and, more to the point, rather temperamental with regard to setting up. Unlike a broad band or linear compander that is insensitive to the frequency characteristics and level setting of the recording chain, Dolby A treats the different bands separately, and therefore level setting can drastically affect the frequency response.
The dbx noise reduction system is an example of a linear compander, and, compared with the complexities of Dolby A, it really is very straightforward, as you can see from the block diagram in Figure 1. On entering the compressor section, the signal is band-pass filtered between 30 Hz and 100 kHz. The low cut-off removes sub-sonic signals from record warps, basso profundo grunts, and so on, that, lying on the edge of record-ability, may well cause ghost modulation of the expanded signal when it emerges off-tape without these deep bass components. The signal then passes through a pre-emphasis circuit that boosts frequencies by 12 dB/octave up to 60 dB. This simple device, in combination with a corresponding de-emphasis circuit in the expander, helps to eliminate modulation noise at high frequencies.
The VCA following the pre-emphasis has a remarkable 120 dB of dynamic range (better even than the Aphex 1537 VCA chip) and may provide the clue to why dbx works as well as it does. This VCA reduces the dynamic range of the input signal by half, subjecting it to 2:1 compression uniformly over the entire audio spectrum. The control voltage for the VCA is derived from a level sensing circuit comprising three further circuits. The first filters the signal emerging from the VCA to encompass a range of frequencies that can be expected to pass fairly flatly through the recording chain. This makes good sense, as there's no point in trying to derive control voltages from frequencies that somehow got left behind! After this, there's a further pre-emphasis stage (this time, 20 dB/octave above 200 Hz for 60 dB) prior to an RMS detector and full-wave rectifier that converts the audio signal into a control voltage for the VCA.
One problem associated with the use of pre-emphasis to limit modulation noise is that it also has the side-effect of causing high frequency saturation of the tape. The additional pre-emphasis circuit prior to the RMS detector circumvents this, so that with high frequencies the RMS detector is put into overdrive and therefore decreases the gain of the VCA to ensure constant amplitude output over the audio spectrum.
When the off-tape signal enters the expander side of the dbx system it first passes through a low-pass filter with a cut-off of 100 kHz to prevent supersonic frequencies generated by bias oscillators and so on from confusing the expansion process. The control voltage is generated by the chain of the same band-pass filter used in the compressor, a de-emphasis circuit to increase the gain of the VCA at high frequencies, and, finally, the RMS detector that delivers the goods to the VCA. This time, of course, the VCA is doubling the dynamic range of the input signal by subjecting it to 1:2 expansion over the audio spectrum, or, at least, that which has successfully emerged from its mauling at the hands (heads) of the tape machine. Finally, there's a de-emphasis circuit on the output of the VCA that restores the highs to their rightful place in the sonic hierarchy.
So, what do we get after putting our precious music through this Krypton factor-like obstacle course of filters? Well, in noise reduction terms, an improvement of 30 to 35 dB in the S/N ratio, which is the practical limit for 2:1/1:2 compansion systems, and, if we're reasonably careful in using the system, a remarkably clean sound that doesn't show too many of the side-effects encountered with the previous generation of dbx. This sounds too good to be true, and, when you consider the reasonable pricing of the Recording Technology Series of dbx noise reduction (£120 for the switchable record/play model 222, £150 for the simultaneous encode/decode model 224), one wonders what sort of reaction the audiophile community will give to it.
Inevitably, the less than enthusiastic press response to the previous incarnation of dbx will cloud the issue, and there's always the nagging suspicion that a compansion system offering so much noise reduction trades something off for giving so much. In practice, everything is fine if (a) the original signal is noise-free, and therefore doesn't contain noise that can be modulated, (b) the heads on the tape machine are accurately aligned and the tape itself is free from dropouts, and (c) the frequency response of the compansion process matches the frequency response of the recording process.
In a studio, all three requirements are routinely met, courtesy of the engineer lining up a multi-track before use and setting levels on- and off-tape, and musicians playing noise-free instruments, or, alternatively, making sure that noise-prone ones are passed through a noise gate.
In a domestic environment, though, it is incredibly difficult to meet these requirements. One of the main bugbears of consumer cassette decks is the remarkable laxity with which manufacturers appear to treat head alignment. Furthermore, the vast range of cassette types available have an equally vast range of frequency responses which makes it impossible to rely upon the off-tape signal having the same frequency spectrum as that coming out of the compressor onto the tape. Add to this quality control problems, in the form of tape dropouts, and you get some idea of the hazards facing the implementation of dbx noise reduction in the average cassette deck-based Hi-Fi system.
With a bit of care, though, dbx, like any compansion system, can be made to acquit itself favourably in a domestic situation, though obviously there are fewer spectral barriers in the way if a reel-to-reel is used rather than a cassette deck. At the recent dbx press launch, various musical excerpts were played to demonstrate how the system performs with both cassette and reel-to-reel machines. The reel-to-reel (a Technics 4-track deck running at 7½ ips) produced some impressive results with no audible squashing of transients or breathing. The cassette machine (a new Sony model), on the other hand, seemed to fare rather less well, and, apart from the fact that the sound just seemed uncomfortable, there was some audible breathing going on. In fact, there's a good chance that this less than impressive performance was due to a mis-aligned deck, but the choice of KEF 105.2s as the speakers for the demonstration didn't help either, as the bass just seemed to roll around on the floor without any real definition.
Quite frankly, I was itching to hear an A/B comparison of the dbx system just switched in and out of a signal's path, but, even though frequency analysis was displayed on the large TV screens (see photo) either side of Jerome E. Ruzicka, the enthusiastic Vice-President of dbx Inc., no attempt was made to use this facility to show the 'before' and 'after' spectral image of music passing through a dbx unit. To quote Mr Ruzicka: 'Nothing is perfect in life, and certainly no companding system is perfect, but, from my view though, any trade-offs involved are very much in favour of hearing this kind of reproduction in the home, with the virtual absence of any background noise'.
So, what it all boils down to is intelligent use. If the user is prepared to take some care in using dbx, then he should be assured of the best that it can offer, and this can be very, very good.
One side of dbx's endeavours that really is 'Very, very good', and something that's reproducible without any additional effort on the part of the listener, is their dbx-encoded discs. Given that digitally-encoded discs are still some years away, and that record manufacturers seem increasingly unable to control the quality of pressings, companded discs seem to offer the only solution to the frustrations of pops and crackles. Apart from the absence of surface noise, these discs also deliver an extraordinary 75 dB of dynamic range when the encoded music is from a conventional analogue master tape; with digital masters, that figure actually comes close to 90 dB on dbx-decoding!
It's very easy to get more excited by such technical specifications than the music itself, and my first reaction to dbx's claim that such encoded discs represent 'the full dynamic range of a live performance' was that they were getting carried away with their own superlatives. I think there is a degree of truth in that pre-judgement, but only with regard to the quality of the discs; the process itself seems in the clear.
I tried a particularly nasty test on the efficiency of the disc decoding by recording some un-decoded music onto a TDK SA cassette with a mid-priced cassette deck (but with heads properly aligned and cleaned!) and then decoding the cassette playback with dbx's disc decoder. Applying this treatment to some digitally-recorded discs of the Tchaikovsky 'Romeo and Juliet Fantasy Overture' and a compilation album of 'Digital Space' (sic) featuring the LSO (yet again) in full flight, the results were truly stunning, with no trace of compansion side-effects, even after being transferred to cassette. The Tchaikovsky was musically the most satisfying of the records I tried out; 'Digital Space' seemed paradoxically earthbound with some uninspired arrangements and a feeling that dynamics were being milked for all their worth just to satisfy dbx.
To play any of the 130 or so dbx-encoded discs currently available (ranging in price from £6 to £12), a disc decoder is needed. The good news is that this is really cheap. BSR Ltd. (the British owners of dbx Inc.) now offer the Model 21 Disc Decoder for around £65. Carping apart, dbx-encoded discs offer some remarkable sonic experiences, and, for rock music, must be near to listening in the JBL-equipped control room of a recording studio.
Dbx's range of products encompasses just about every audio application that can be made of compressors and expanders, and includes dynamic range expanders, dynamic range enhancers, their professional series of noise reduction and compressor/limiters, together with the extraordinary subharmonic synthesiser, or 'boom box', that generates gut-crunching infrasonics.
A recent addition to their range, the 20/20 Computerized Equalizer/Analyzer, looks particularly interesting. This unit combines a microprocessor-controlled 10-band graphic equaliser, real-time analyser, pink noise generator, sound pressure level meter and a calibrated microphone. By setting the mic at a desired listening position, starting up the pink noise, and pressing the AUTO EQ button, the 20/20 performs a spectral analysis of the sound reaching that location from the speakers of your Hi-Fi system and automatically establishes the EQ curve for a flat response.
In addition, to quote dbx: 'Because the 20/20 has ten memory locations, you can equalise each of your favourite listening positions, then store them for recalling later at the touch of a button'. At a price getting on for four figures, the 20/20 is definitely a rich man's toy, but, with this addition to the dbx range, I'm beginning to wonder how much further this quest for perfection will go. Perhaps dbx's next step will be to standardise the process of composition itself!
As long as society (with some glaring exceptions) allows Man to perform and experience different types of music, and as long as there's no standardised accuracy of auditory perception in a given individual, then people will continue to argue on the merits and demerits of heavy turntable platters, digital recording, and noise reduction systems like dbx. Perhaps, now, we could get back to the task of helping musicians and composers create the music without which dbx would be out of a job!
Gear in this article:
Review by David Ellis
mu:zines is the result of thousands of hours of effort, and will require many thousands more going forward to reach our goals of getting all this content online.
If you value this resource, you can support this project - it really helps!