The Suppression of Surface Noise, by D.T.N. Williamson--A classic which laid down constraints for all today's quieting systems.
[A resume of a lecture given at the B.S.R.A. Convention on May 15, 1953, and reprinted from the Journal of the British Sound Recording Association. ]
NOISE ON GRAMOPHONE records can be divided into two main classes: that resulting from the granular structure of the record material, and that arising from dust and surface blemishes. The former produces a steady hiss of approximately ''white'' noise which is found mainly on 78 r.p.m. records. The latter is impulsive in nature and is of primary importance in microgroove recordings because the size of an average dust particle is greater relative to groove size.
========
---- (left) "Sticky-roller" cleaner does not reach deep enough to
clean groove bottoms (1,000x). (right) Fibers of velvet record cleaner as they
sweep the record grooves (1,000x).
============
A pickup which has a narrow band width tends to integrate noise energy outside its passband. This energy appears in the form of ringing near the cut-off frequency. Resonances within the passband produce the same effect. This ringing is readily detectable and has a more unpleasant effect on the listener than the same total noise energy evenly distributed. Impulse noise consists almost entirely of ringing at the cut-off and resonance frequencies unless the rate of change of amplitudes at these points is low.
The shape of the stylus has an important bearing on the amount and character of surface noise, especially impulse noise. The noise energy is proportional to the radius of curvature of the section of stylus in contact with the groove, and experiments show that with 78 r.p.m. discs the steady noise level can be reduced by 3-6 decibels by the use of an elliptical stylus with a radius of curvature at the contact point of 0.6 mils, compared with the more usual spherical stylus radius of 2.5 mils. The signal / noise ratio at high frequencies is, of course, also greatly improved.
Various systems have been devised to reduce the audible effect of surface hiss, the high-frequency components of which are the more offensive. The impulse interference is not normally suppressed by these devices.
An obvious method of reducing the effect of the noise is, of course, to reduce the frequency range and, in fact, this is the most common remedy. It is, however, a very unsatisfactory one, as apart from the obvious disadvantages, the integration of pulse noise does little to reduce its offensiveness, but merely redistributes the energy in the lower frequency band.
The problem, therefore, is to remove an interfering signal which is essentially of similar character to the recorded music and is intermingled with it. To do this it is necessary to find some distinguishing feature which will enable the unwanted signal to be recognized and removed. The usually adopted distinguishing feature is the low and relatively constant amplitude of the noise compared with the signal.
That this is not a rigidly definable distinguishing feature is obvious, and to this must be attributed the frequently unsatisfactory behavior of most noise Suppressors.
The ''vertical'' type of noise suppressor due to Olson, so called because it discriminates on a basis of amplitude, is a very interesting and elegant one. This type of suppressor functions by suppressing entirely the signal below a certain level, this level being set just above the mean noise level. This is an interesting process because, although it involves considerable amplitude distortion, it is theoretically possible to ensure that this distortion does not appear in the output.
In practice, the signal spectrum is split into two or more bands, the upper of which is one octave wide. These bands are dealt with separately and then recombined. Considering a two-band system, the lower section might extend from 20 to, say, 6,000 cycles per second and the upper from 6,000 to 12,000 cycles per second.
The signal is applied to a group of filters, a low-pass with a cut-off frequency of 6,000 cycles per second, and a band pass, filtering from 6,000 to 12,000 cycles per second. Most of the audibly offensive noise is concentrated in the upper band, which also contains the harmonic energy essential for accurate reproduction. The signal in the upper band is applied to a non-linear network, with a transfer characteristic such that all signals below a specified level are sup pressed.
Since all signals going through the band-pass filter are of sinusoidal form, as it is only an octave wide, the output from the non-linear network has a discontinuous character near zero level. This, of course, represents a very obnoxious form of distortion and the signal could not be utilized in this form. If, however, it is passed through another similar band-pass filter, its sinusoidal form will be restored, because of harmonic suppression.
Turning now to the pulse type of noise peculiar to microgroove recordings, it is found that the duration of the pulse usually lies between 25 and 150 microseconds.
Examination of the Fourier transform of such a pulse shows that the energy spectrum is divided into a series of groups with minima at intervals determined by the length and shape of the pulse. The energy in each group diminishes with increasing frequency. The spectrum of an actual pulse extends as far as the response / frequency characteristic of the pickup and the rest of the equipment permits. On the other hand, the spectrum of the recorded music is usually limited.
This may be due to the use of filters in the recording equipment or to the limited frequency range of recording heads, or possibly to the erasing effect of the bevel angle of the recording cutter.
If, therefore, a pickup with a frequency range much wider than that which is recorded is used, energy due to a speck of dust should be discernible in the section of the spectrum above that of the music.
This, in fact, proves to be the case, and although occasionally on some recordings musical energy appears at frequencies as high as 100 kilocycles per second, mainly during cymbal clashes, these occasions are rare.
We have, therefore, a means of determining the presence of a pulse, the indication being clear of the music except on rare occasions. In practice, a high-pass filter with a cut-off frequency around 20 kilocycles per second separates the high frequency components of the pulse fairly adequately from the music, although it may be necessary to raise this frequency to about 40 kilocycles per second if complete separation is necessary. The signal in this channel may be used to control some circuit which removes the pulse.
The shape and duration of the pulse, of course, determines the precise nature of the waveform in the upper channel and its relation to the complete pulse. A pulse with a very sharp leading edge has high frequency components which produce in the upper channel a waveform with a similarly sharp edge, coincident with that of the pulse. Pulses of longer duration with a slower leading edge produce a waveform in the upper channel, the peak of which may be delayed by a significant amount with reference to the leading edge of the pulse.
The principle of using the pulse to gate itself is based on the presumption, experimentally verified, that it is possible under certain conditions to shut a musical channel down for a period of one quarter of a millisecond without introducing audible distortion. It seems, in fact, to be possible to remove up to about one tenth of the music in any given period before any effects become noticeable. Provided the gating is random, however, it is difficult to detect.
The criterion of effectiveness is that the gating action must produce a smaller disturbance than the one which it is in tended to remove. Since we are intending to gate a pulse produced by a speck of dust, during which the stylus partly loses contact with the groove and there is unlikely to be any musical energy present, we would in theory not be losing anything useful if we shut down the channel for its duration.
The simplest method of suppression would be to use the pulse in the upper channel, suitably shaped, to gate or cancel, itself in the main channel, but this is rendered impracticable by the uncertainty of the delay, and the best alternative seems to be to generate a gating pulse which is sufficiently long to straddle the interfering pulse.
Measurement shows that the time between the occurrence of the leading edge of the pulse and the appearance of the trigger pulse in the upper channel varies between zero and about 150 microseconds, and therefore a delay of this order must be introduced in the signal channel. This is most conveniently done by means of a delay line.
It is not usually realized that it is possible to delay an audio frequency by quite an appreciable amount very simply.
Delays of several milliseconds can be obtained quite practicably by using passive delay networks. In this case, the introduction of a delay of 200 microseconds involves only the use of a simple coupled inductor and a number of capacitors.
The most usual type of gate is one which reduces the signal level to zero for a period, the signal then being restored to the value which it would have held at that instant in time, had it not been interrupted. This means that the waveform falls sharply to zero and then rises sharply to its new value, producing a pulse, the mean amplitude of which is equal to the instantaneous amplitude of the waveform which is being gated.
If the waveform happened to be zero at the instant of gating no pulse would occur.
The gate thus substitutes for the pulse that it removes, one of similar duration which has a height not exceeding the peak signal amplitude. This will only provide an improvement if the interfering pulse amplitude is larger than the peak signal amplitude, but this, of course, is the normal state of affairs where suppression is worthwhile.
It is possible to envisage another type of gate which leaves the waveform at the instantaneous level or ''previous value'' at which the gate operated, instead of reducing it to zero. This causes less disturbance of the waveform than the previous type and its operation is less audible. An equivalent circuit consists of a low impedance signal source connected through a switch to an R.C. network with a long time-constant, the impedance of the network being much higher than the source impedance.
The disturbance caused by a gate operating on this principle depends very much on the frequencies being gated.
When gating a pulse, it is essential that the action occurs before any rise takes place, as otherwise the gating action will substitute a rectangular pulse of the duration of the gating waveform.
An interesting possibility is that, in stead of reducing the signal to zero for the duration of the pulse, another pulse-free section of the signal immediately following the pulse may be substituted for the pulse mangled section, this procedure being based on the supposition that, on a statistical basis, a section of music as short as, say, 400 microseconds (being twice the length of a pulse) is not likely to change very much.
The procedure then would be to allow the pulse to enter the delay line and, just as it was due to emerge, to cheat it by switching the beginning of the line until the pulse had cleared off, when the channel would be restored to the end of the line, just in time to hear the same 200 microseconds' worth of signal again! The switching is, of course, carried out automatically by the signal in the upper channel, and disturbance could be minimized by producing a smooth changeover.
Before any of the methods of pulse suppression outlined above can be considered entirely satisfactory some improvement in the technique of gating is required, aimed at producing minimum disturbance in the wanted signal. The development of a satisfactory system of microgroove impulse noise suppression is imperative if disk recordings with their admitted advantages of swift duplication at low cost are to survive for long beside the rapidly developing and comparatively noiseless process of magnetic tape recording.
-----
Also see:
Direct Metal Mastering-- A New Art In LP Records (April 1987)
The Advent MPR-1 Microphone Preamp, by Frederick M. Gloeckler, Jr. --A two-channel, transformer input, control-less amp for Advent's 201A