In this exclusive preview, Paul White is introduced to a brand‑new way of making high‑resolution digital recordings.
Over the past few years, digital recording has become the standard, both in professional and private studios, but already the limitations of the conventional digital format are beginning to show. As we demand more resolution, the technical problems associated with designing and building converters that can discriminate voltage steps of less than one microvolt threaten to call a halt on what can be achieved, and as calls for higher sample rates are heeded, these problems will get worse. There's also the problem of data corruption. No matter how powerful the error‑correcting system, there comes a point where the data loss is too great to reconstruct, and the result is a glitch. Anyone with more than a little experience of DAT recorders will be familiar with the situation where a tape made on one machine glitches badly on another — a problem caused by the need to read data from precise areas of the tape's surface at precise times.
Having had several years to study the pros and cons of various systems, an ARS working group, in co‑operation with data‑recording specialists LOG, have come up with a different approach that is surprisingly robust, and also avoids weaknesses such as quantising noise and sample‑rate jitter. Because tape is still the lowest‑cost medium, preliminary trials using a stereo machine have been carried out to test the validity of the process, and so far the results are encouraging. Indeed, we could see commercial recorders based on this technology being introduced into the marketplace within the next two years; given the professionals' dissatisfaction with DAT, I anticipate a lot of interest.
Sample Rate
Much has been said recently of the need for higher sample rates but, as mentioned earlier, the faster a converter runs, the more difficult it is to retain accuracy, and even the very best 24‑bit converters running at 44.1kHz don't manage much in excess of true 21‑bit dynamic range. This new system does away with conventional converters altogether — instead, the sample clock, which is sinusoidal rather than a square wave, is simply mixed with the analogue input signal; the result is then applied to a fixed magnetic recording head where a magnetic flux proportional to the composite signal is transferred to a section of moving tape. The sample clock can be anything from 60kHz to 100kHz, and the analogue input must be band‑limited, as in all digital systems, so that the maximum input frequency never exceeds the Nyquist limit of half the sample rate. The beauty of this system is that, instead of being limited to a fixed number of discrete quantised steps, its quantisation is carried out at molecular level by the amount of energy required to flip each magnetic molecule from one state of its hysteresis curve to the other. In effect, each magnetic molecule stores one bit of data.
Using this basically simple approach, recordings can be made onto very narrow bands of tape, and the current systems use a split head so as to record stereo signals as two discrete tracks. Of course, problems do arise because of the statistical nature of tape — if everything was perfect, you'd be able to record onto a track comprising a single line of oxide molecules, but, in reality, flaws in the tape coating and misalignment of the oxide molecules mean that such a recording would be flawed by errors. However, the errors would show up as statistical noise rather than as glitches, which gave the designers an idea of how to tackle the problem.
Error Correction
Conventional error correction works by recording a certain amount of redundant data, then using powerful algorithms to detect and then correct faulty blocks of information. In contrast, this new system simply uses a slot‑shaped gap in the record head so as to record the same data over a wider strip of tape. When replayed, the data is, in effect, read back several times from a huge number of parallel, one‑molecule‑wide tracks and then averaged. This way the error correction relies on natural statistics rather than mathematical codes. Because correlated signals sum linearly (uncorrelated noise both adds and cancels), using this form of error correction results in a robust signal with a high degree of integrity and a high resistance to data failure. If a surface oxide flaw is encountered, the statistical summation process will be compromised, but when the signal is decoded, the outcome is simply a momentary increase in noise or a drop in level rather than a complete data loss.
...quantisation is carried out at molecular level by the amount of energy required to flip each magnetic molecule from one state of its hysteresis curve to the other.
Perhaps a greater benefit of the system is that, because conventional converters are not employed, there is no quantisation noise other than that created by the statistical nature of the magnetic domains on the tape's surface. This statistical noise sums to produce near‑Gaussian broad‑band hiss — in effect a perfectly implemented dither system at quantum level. Furthermore, because the sample clock is sinusoidal, and because the oxide molecules are randomly spaced on tape, the playback sampling rate is essentially random and determined by the rate at which individual oxide molecules pass over the replay head.
Decoding
The recorded signal is picked off tape using a stationary magnetic head in the conventional way. No buffering is used either at the record or playback stage, and conversion back to analogue requires only that the the sample clock be excluded from the output signal by means of filters. This produces a smooth analogue output which can be resampled with no fear of clock clash. The lack of buffering does make the system subject to pitch modulations introduced by motor‑speed errors, but the latest generation of quartz‑locked direct‑drive motors reduces this to an insignificant level. Formatting the tape simply requires the oxide molecules' magnetic state to be randomised by the application of the raw sample clock to a separate head positioned a short way before the record/replay head. In this way, the tape can be formatted as it is recorded — there's no need to waste time striping the tape prior to a session.
Because there's no hard clocking or dividing of data into packets, tape speed may be varied for traditional varispeed effects without affecting the data integrity, and it's even possible to turn the tape over for reverse effect — something conventional digital machines can only dream of. The best news yet is that, because of the simple mechanics of the system, it may be possible to adapt old open‑reel machines at a relatively low cost. What's more, using conventional quarter‑inch tape retains the subtle harmonic distortion characteristics of this medium, which means that getting a vintage sound is as easy as threading up a reel of tape.
Future plans include a multitrack version of the machine based around a wider tape format, possibly 50.8mm, and a multi‑faceted record head capable of recording a number of discrete tracks side by side. It's also possible that the dynamic range of such a recorder may be further extended by applying frequency‑ and level‑sensitive compression to the input signal and then applying a reverse function to the playback audio directly after conversion.
The first production machine will be a stereo mastering recorder built under the LOG badge, and entitled the Advanced Non‑coding Audio LOG recorder — ANALOG for short. From what we've heard so far, it has a great future.