by H. W. Hellyer [ Bristol, England]
Fig. 1--A simple muting circuit used by Panasonic-simple, but effective,
sensing the signal level and "killing" the line output when
the signal drops dangerously near the noise level. The circuit shown
is for one channel. The same two transistor network is employed for the
other channel, and this "commoning" can lead to problems.
LET'S TAKE A LOOK at one or two ventures into noise exclusion that have been at least a bit more ambitious than a mere clipping of playback peaks. One such system is Panasonic's NFD device. NFD, quite simply, mutes the line output unless the signals (on playback) are above a predetermined level and below a set frequency. This reduces hiss when the signal level is low. That is, you get what you want when you most want it.
In the RS 735US, there was a two-transistor, nine-diode circuit that gave very good results indeed. Figure 1 shows the basic configuration. Signal-to-noise ratio, when I tested it, with this noise filter employed, was as good as 66.5 dB. At 1 kHz, the improvement was a mere 3/4 dB, but although at rated output level the NFD only made 1 dB difference to the S/N ratio, when the level of signal was down around the danger level, approaching what would have been obtrusive hiss, the circuit effectively blanked signal, and its action did not, as with so many compandor systems, provide an aural switchback.
Taking the replay system a step farther, Philips has the DNL innovation, which should make much cassette work with other folk's tapes a really feasible possibility.
DNL means Dynamic Noise Limiter, and Philips (Norelco to you) argues thus .. . "When music is played softly, it is made up almost entirely of pure tones in the middle and low frequency ranges with hardly any harmonics. This is mainly because very few musical instruments produce tones whose fundamental frequencies are much higher than 4.5 kHz. Tape hiss, however, is made up of sounds in the higher frequencies so that it is during the soft passages and silent intervals that it becomes most noticeable.
"When music is played loudly, it not only contains the lower and middle frequency pure tones, but also a great deal of harmonics, which give character to the sound. It is in the loud passages that noise suppression is unnecessary as the high frequency harmonics hide the tape hiss. Any filter action would make the music sound dull and unnatural.
"Therefore, if tape noise or hiss is to be suppressed, it must be completely eliminated in periods of no music signal, reduced during the soft passages of music, and left unsuppressed during the loud passages." Thus, the oracle-begging one or two questions, like: "Pure tones--all instruments played softly?" and "What happens to the soft tones of one instrument when another plays loudly?" and "How soon after the loud noise ends does the suppression take place?" The Dynamic Noise Limiter acts on replay, the argument being that it therefore allows complete compatibility, giving the benefit of noise suppression even to those poor, deprived owners of untailored cassettes. It is, effectively, a steep, lowpass filter which acts when there are no high signal frequencies.
Philips has been rather clever about it, allowing high frequency signals that exceed a predetermined level to bypass the filter: so there are two signal chains. Fig. 2 shows the block diagram. From the splitter, the signal takes two paths, one path merely inverts the phase without affecting the linearity while the other passes it through the tailoring process.
This process chops off the lower and middle frequencies, leaving only those above 4 kHz (approximately-you can't do these chopping actions abruptly without introducing almost ineradicable distortions, whatever the advertising copywriters say). This remaining high frequency band is now monitored so that the quieter parts of higher frequency are boosted. Hence the variable attenuator--it is both level and frequency-conscious.
Fig. 2--Block diagram of the Philips (Norelco) Dynamic Noise Filter.
The surprisingly effective though unsophisticated system acts on playback
only and has the effect of an 18 dB/octave filter when the signal is
low. A S/N ratio improvement of around 10 dB at 6 kHz and 20 dB at 10
kHz has been measured (unweighted). The high-pass filter takes effect
above 4 kHz.
Fig. 3--The DNL circuit, four transistors, six diodes, and a handful
of common components, can easily be made up into a neat set-side box-no
bigger than a double pack of 20 cancer-sticks.
Adding together the processed and unprocessed chains should now, theoretically, give a signal whose low-level high frequencies have a quietened effect, while middle and low frequencies are unaltered and where the higher volume high frequencies are given their full, required weight. In theory, once again, the result should be a true replica of the original, but without the hiss.
And, I must admit, despite some initial misgivings because Philips demonstrated this device to us a year or so ago in an hotel room whose air-conditioning added some 30 dB to the ambient noise, the subjective effect is a cleaner sound, whatever the condition of the recording.
But I still feel that the answer is not to use a circuit that gives, as Philips claim, a 10 dB improvement of S/N ratio at 6 kHz and a 20 dB improvement at 10 kHz on replay, but to improve the overall record/replay process in such a way as to retain its original sound structure, not "tailor" it. Again, if you must have slow-speed, narrow-track recording, then you have to engineer out the hiss, not allow it to happen and then try to beat it.
So we come to Dolby and the now-famous stretching process that Dr. Ray Dolby pioneered. The original "A" process aimed at beating the "breathing" that compansion procedures forced disc users to suffer and cost more than some recording companies could afford. It begins its work during recording, splitting the audio path into a direct and a rectifier chain. But the expensive "A" system did this in four bites, carving up the frequency spectrum to give differential gain depending on signal level within the frequency bands. These are: below 80 Hz, from 80 to 3,000 Hz, above 3,000 and again above 9,000 Hz.
Both hiss and hum are present in the recording process, and while hum can be relegated to one low portion of the audio spectrum, hiss is a very different problem. It obtrudes into the very region where our ears happen to be most sensitive. It has measurable components that extend way upwards into what some engineer colleagues of mine call the "annoyance pass band." Any crude way of militating against hiss will mutilate the upper frequencies which we need to preserve to get the clash and tingle of a full musical experience.
Fig. 4--One way of explaining the Dolby system: The original signal
has its lower, levels down around the system noise. Processing during
record gains some 10 dB of S/N ratio. Replay retains this, raising the
lower levels of signal that much above the noise.
Fig. 5--An alternative explanation, as depicted by Dolby: A, music is
made of sounds of different loudness with intervals of silence; B, noise
of some kind is inescapable; C, when a tape recording is made and replayed
the noise interferes with the low level signals, spoiling the program;
D, the Dolby system boosts the lower signals during recording; E, those
lower signals are still above the annoying noise during replay, as shown
in F, the composite picture of the reconstituted sound with noise "reduced" by
the carefully engineered boost and stretch system.
Fig. 6--Noise reduction units can be added quite easily to existing
equipment. This Advent 100A has been enthusiastically received, despite
the $250.00 price tag. My own special interest is harmonic distortion,
and I was interested to note that the 100A was under 0.4% to 0 dB and
less than 0.2% at lower levels. Output noise, -60 dB; noise reduction
around 10 dB above 4 kHz, about 3 dB at 6 kHz. This is a stereo unit
and well worth considering for slow-speed recording.
Again, the procedure is to let the noise remain when the music is loud enough to mask it. Masking--as a technical term--is a peculiar business. It depends as much on relative frequencies as on loudness, and has some strange anomalies to do with time difference and phase factors. Subject for a later discourse, maybe. At present, please take my word for it that the phenomenon happens, and by letting the main, high level signals straight through the system, Dr. Dolby follows the method we have roughly outlined already.
The subtlety lies in the treatment of the low-level signals, where noise is obtrusive. Dolby calls this the differential component, and this is, of course, relatively small-and hence more difficult to handle. It has to be remembered that the noise reduction system does not eradicate noise; it boosts weak signals to improve the signal-to-noise ratio, that's all.
That's all! Pause for hollow laughter! Arguable decisions are the threshold limit, below which noise-plus-signal will be processed, attack time, the response of filter circuitry to the information that a signal in need of treatment is coming along, the amount and nature of compression, and the way of ensuring a mirror image expansion and an avoidance of overshoot (which would process signals that did not need such treatment). If the distortion has a duration of less than a millisecond, it will defeat the human ear. This is a smaller fraction than normal signal transients and our aural loudness-growth characteristic cannot distinguish the short-lived distortion.
The Dolby "B" system came into being when Ray Dolby was asked to dream up a modified noise reduction device for use with domestic equipment. The only feasible way to keep such a system within our budget was to forgo the technical requirement of four passbands and operate over the whole audio spectrum, this time making the sensor part of the apparatus listen for frequency as well as loudness, on a kind of sliding scale.
The system comes into action at about 600 Hz, with a maximum 3 dB effect. (O.K., so the ads say it extends above 2 kHz, but the sliding scale method means it really begins lower down). At 1.2 kHz it has a maximum 6 dB effect, has 9 dB at 2.4 kHz and reaches the advertised 10 dB above 4 kHz. The compression comes in about 45 dB below what has become known as the "Dolby level." This can be defined as a flux level on tape of 200 nano-webers per meter. Call this 0 VU. In more technical terms, the differential chain splits into the rectifier path and into the linear path to the mixer for re-addition to the main signal. The rectifier path contains boost circuits giving a 6 dB per octave flip to the higher frequencies.
Then the output is rectified. This rectified signal effectively alters the dynamic resistance of an FET at the input end of the chain, and so gives a boost at low dynamic levels and practically no boost at high levels. By the simple device of driving the FET via a small coupling capacitor, Dr. Dolby achieved both a drop in gain with an increase in dynamic level and a change of the turnover frequency of the "threshold" as the level changes. The sliding scale, in fact.
At low levels the capacitor lets the FET see the full signal and gives a good 10 dB boost above 2 kHz. Increase the input level and the frequency above which this full boost is given begins to rise. Turn up the wick still more and the treble boost in the rectifier chain stops the over-saturation of the tape. To reinterpret, that means the tape is driven to its full limit when need be, at high dynamic levels (of original signal), but is allowed up to a 10 dB boost at lower signal levels. The replay mode is reciprocal.
The entire processed chain is inserted in a feedback loop around the main chain to subtract instead of add. The elegance of the system is that the same basic circuitry, and, indeed, a mirror-image printed circuit board makes production costs tumble and the add-on Dolby units now available should be within any enthusiast's purse. (Dolby IC chips are also coming soon. -Ed.)
Fig. 7--Slim, elegant, technically precise, one section of the Dolby
A system as used by professional recording bodies throughout the world.
Having had the chance to "rip one to bits," I can vouch for
its engineering excellence.
Fig. 8--Block diagram of the Burwen system, with refinements like active
transformers and direct play equalizers omitted.
The heart of the system is the rectifier module, monitoring the gain of two channels simultaneously in the "domestic" system. Operational amplifiers are used widely in this system with very high accuracy as a result.
My own tests with those available in the U.K. have confirmed that signal-processing of cassette-recorded music, speech, and sound effects have done wonders to guard against hiss and have not made detectable any audible worsening of the prime signal.
After Dolby, what? Well, according to Richard Burwen, quite a lot. In the December, 1971 issue of the Journal of the Audio Engineering Society, I came across the Design of a Noise Eliminator System which gave me much brain-searching and is at present exercising the pundits in those polite tomahawkeries of the erudite correspondence columns. (See also AUDIO, June, 1971.) To begin with, the title of Richard S. Burwen's paper hits a sore point. The only way you eliminate noise, truly, is not to cause it. After the die is cast, all you can do is guard against it-which we have seen three different systems doing in the preceding notes.
Mr. Burwen took the critics by the ears at the 41st convention in New York on October 5, 1971. In February of that year, a paper of his entitled "A Dynamic Noise Filter" had aroused comment. He is more concerned with studio tape machines, just as Dr. Dolby was, and there seems little hope, at present, of such an elegant "domestic" solution to the noise reduction problem with a plain man's Burwen. But anyone who has been in the audio field as long as us (well, me) knows better than to say that something, anything, cannot be done.
So let's conclude with a brief look at Mr. Burwen's solution. He set himself some pretty high parameters. His system was not, he told us, to exceed the present 1%, and preferably 0.5%, distortion level of good taping. He wants to record live music "with no audible noise whatsoever." So his first experiments were to determine peak recording levels.
Recording to +3 VU, a normal process, when 0 VU is the standard set limit and peaks above this as much as +6 VU are occasionally tolerated because of their short duration, meant that distortion on tape went over that critical 1%. He concluded--first point, and first stumbling block for his critics--that it is not always advisable to retain every peak.
Listening tests revealed that for noise to be negligible in the absence of program material, it had to be 90 dB or more below the 1% distortion level, i.e., better than -84 VU. Then he found that noise 65 dB down was audible with a 500 Hz sine wave but masked by frequencies above 3 kHz. You could reduce the bandwidth to about a half-octave centered on 500 Hz and get a pure tone-so the solution seemed to be split the waveband, per Dolby A. But the multiband system, according to Richard Burwen, has the disadvantage of frequency response errors in the tape machine causing errors in the expansion process. The solution was to use the whole band but compress the 90 dB expected input to 30 dB at the tape. He then combined the principles of his dynamic noise filter (see June, 1972 issue of this magazine) with a single wideband compandor.
The dynamic noise filter acts as a low level expander at top and bottom of the frequency spectrum--again, something like we've seen before. Adding a high and low-frequency compression system seemed to be the answer, and high frequency pre-emphasis was intended to improve the S/N ratio. Some hellish problems raised themselves at this point, and Mr. Burwen went back to the drawing board. He finally produced three systems, A, B and C. Characteristic A is optimized for studio recording at 15 ips. It has a dynamic range of 110 dB and this is the one you'll see hailed in the ads! System B operates more modestly to give a 102 dB dynamic range at 7 1/2 ips, and C is the one that may eventually interest us at 3 3/4 or 1 7/8 ips for FM broadcasting, records or background music. If you want it in the words of the master: "The system ... utilizes high and low frequency pre-emphasis and a single wideband cube root compressor to produce the recorded signal, and a complementary expander and pre-emphasis for playback." The important point slipped in later is that in the single band system the frequency response is constant and is not affected by inaccuracy in the tape machine. Again, we shall leave the pundits to argue.
The high performance of the Burwen circuitry has been made possible by the low-noise two-quadrant multiplier/ divider. Bettering Dolby by one magnitude in claim and applicable also to FM systems, it seems to offer possibilities, and we must wait and see what the outcome may be.
For my part, in this noise-polluted environment, I welcome any device that can help rid us of clamor. But noise is what you make it, and the tick of an obtrusive clock, as many an amateur recordist has found, can be as bothersome as a traction engine. The subjective results, applied to cassette, have been enormous-praise to the noise-breakers!
==========
(adapted from Audio magazine, Oct. 1972)
Also see:
A Dynamic Noise Filter for Mastering (June 1972)
Dolby B-Type Noise Reduction System (Sept. 1973)
Dolby B-Type Noise Reduction System--Part 2 (Oct. 1973)
Introducing Dolby S-Type Noise Reduction (Jun. 1990)
EQ & NR: Striking A Balance (Aug. 1988)
Deck to Deck Matching and NR: Straightening the Mirror (Aug. 1986)
= = = =