Chapter 7

Noise reduction

Noise reduction techniques have been applied to analogue tape machines of all formats, radio microphones, radio transmission and reception, land lines, satellite relays, gramophone records, and even some digital tape machines. The general principles of operation will be outlined, followed by a discussion of particular well-known examples. Detailed descriptions of some individual systems are referred to in the Further reading list at the end of this chapter.

Why is noise reduction required?

A noise reduction system, used correctly, reduces the level of unwanted signals introduced in a recording–replay or transmission–reception process (see Figure 7.1). Noise such as hiss, hum, and interference may be introduced, as well as, say, print-through in analogue recording, due to imperfections in the storage or transmission process. In communications, a signal sent along a land line may be prone to interference from various sources, and will therefore emerge with some of this interference signal mixed with it. A signal recorded on a cassette machine replays with high-frequency hiss. Unwanted noise already present in a signal before recording or transmitting, though, is very difficult to remove without also removing a part of the wanted signal. One could roll off the treble of a cassette recording to reduce the hiss, but one would also lose the high-frequency information from the sound, causing it to sound muffled and ‘woolly’.

Methods of reducing noise

Variable pre-emphasis

Pre-emphasis (see Fact File 7.1) is a very straightforward solution to the problem of noise reduction, but is not a panacea. Many sound sources, including music, have a falling energy content at high frequencies, so lower-level HF signals can be boosted to an extent without too much risk of saturating the tape. But tape tends to saturate more easily at HF than at LF (see the previous chapter), so high levels of distortion and compression would result if too much pre-emphasis were applied at the recording stage. What is needed is a circuit which senses the level of the signal on a continuous basis, controlling the degree of pre-emphasis so as to be non-existent at high signal levels but considerable at low signal levels (see Figure 7.2). This can be achieved by incorporating a filter into a side-chain which passes only high-frequency, low-level signals, adding this component into the un-pre-emphasised signal. On replay, a reciprocal de-emphasis circuit could then be used. The lack of noise reduction at high signal levels does not matter, since high-level signals have a masking effect on low-level noise (see Fact File 2.3).

Figure 7.1   Graphical representation of a companding noise reduction process

Figure 7.2   A simple complementary noise reduction system could boost high frequencies at low signal levels during encoding, and cut them on decoding (encoding characteristic shown)

Such a process may be called a compansion process, in other words a process which compresses the dynamic range of a signal during recording and expands it on replay. The variable HF emphasis described above is an example of selective compansion, acting only on a certain band of frequencies. It is most important to notice that the decoding stage is an exact mirror image of the encoding process, and that it is not possible to use one without the other. Recordings not encoded by a noise reduction system cannot simply be passed through a decoder to reduce their noise. Similarly, encoded tapes sound unusual unless properly decoded, normally sounding overbright and with fluctuations in HF level.

Dolby B

The above process is used as the basis for the Dolby B noise reduction system, found in most cassette decks. Specifically, the threshold below which noise reduction comes into play is around 20 dB below a standard magnetic reference level known as ‘Dolby level’ (200 nWbm⁻¹). The maximum HF boost of the Dolby B system is 10 dB above 8 kHz, and therefore a maximum of 10 dB of noise reduction is provided. A high-quality cassette deck, without noise reduction, using a good ferric tape, will yield a signal-to-noise ratio of about 50 dB ref. Dolby level. When Dolby B noise reduction is switched in, the 10 dB improvement brings this up to 60 dB (which is more adequate for good-quality music and speech recording). The quoted improvement is seen when noise is measured according to the CCIR 468-2 weighting curve (see ‘Dynamic range and signal-to-noise ratio’, Appendix 1) and will not be so great when measured unweighted.

Dolby B incorporates a sliding band over which pre-emphasis is applied, such that the frequency above which compansion occurs varies according to the nature of the signal. It may slide as low as 400 Hz. This aims to ensure that maximum masking of low-level noise always occurs, and that high-level signals at low frequencies do not result in ‘noise pumping’ (a phenomenon which arises when a high level signal in one band causes less overall noise reduction, causing the noise in another band to rise temporarily, often not masked by the high-level signal due to the difference in frequency of the signal and the noise).

The Dolby process, being level dependent, requires that the reproduced signal level on decoding is exactly the same with respect to Dolby level as on encoding. This means that a particular cassette machine must be set up internally so that Dolby-encoded tapes recorded on it or other machines will replay into the decoder at the correct electrical level for proper decoding. This is independent of the actual output level of the machine itself, which varies from model to model. If the replay level, for instance, is too low, the decoder applies too much treble cut because the −20 dB threshold level will have moved downwards, causing recorded signal levels above this to be de-emphasised also. Frequency response error will therefore be the result. Similarly, if the frequency response of a cassette machine shows significant errors at HF, these will be exaggerated by the Dolby record/replay process.

A so-called MPX (multiplex) filter is mandatory with Dolby B systems, and removes the 19 kHz pilot tone present in FM stereo radio broadcasts. This is needed because the pilot tone may still be present in the output of an FM tuner, artificially affecting the encoded level of HF signals on a recording from the radio. Since the frequency response of many cassette machines does not extend to 20 kHz the tone would not be reproduced on replay, and thus the decoder would not track the encoded signal correctly, leading to noise pumping and response errors. On some recorders the filter is switchable. On cheaper machines the filter simply rolls off everything above 15 kHz, but on better machines it is a notch at 19 kHz.

Dolby C

Dolby B became widely incorporated into cassette players in the early 1970s, but by the end of the 1970s competition from other companies offering greater levels of noise reduction prompted Dolby to introduce Dolby C, which gives 20 dB of noise reduction. The system acts down to a lower frequency than Dolby B (100 Hz), and incorporates additional circuitry (known as ‘anti-saturation’) which reduces HF tape squashing when high levels of signal are present. Most of the noise reduction action takes place between 1 kHz and 10 kHz, and less action is taken on frequencies above 10 kHz (where noise is less noticeable) in order to desensitise the system to HF response errors from such factors as azimuth misalignment which would otherwise be exaggerated (this is known as ‘spectral skewing’). Dolby C, with its greater compression/expansion ratio compared with Dolby B, will exaggerate tape machine response errors to a correspondingly greater degree, and undecoded Dolby C tapes will sound extremely bright.

Dolby A

Dolby A was introduced in 1965, and is a professional noise reduction system. In essence there is a similarity to the processes described above, but in the Dolby A encoder the noise reduction process is divided into four separate frequency bands, as shown in Figure 7.3. A low-level ‘differential’ component is produced for each band, and the differential side-chain output is then recombined with the main signal. The differential component’s contribution to the total signal depends on the input level, having maximum effect below −40 dB ref. Dolby level (see Figures 7.4(a) and (b)).

Figure 7.3   In the Dolby A system a low-level ‘differential’ signal is added to the main signal during encoding. This differential signal is produced in a side-chain which operates independently on four frequency bands. The differential signal is later subtracted during decoding

Figure 7.4   (a) Differential signal component produced in a Dolby A side-chain. (b) Input level plotted against output level of Dolby A unit after adding or subtracting differential component

The band splitting means that each band acts independently, such that a high-level signal in one band does not cause a lessening of noise reduction effort in another low-level band, thus maintaining maximum effectiveness with a wide range of programme material. The two upper bands are high pass and overlap, offering noise reduction of 10 dB up to around 5 kHz, rising to 15 dB at the upper end of the spectrum.

The decoder is the mirror image of the encoder, except that the differential signal produced by the sidechain is now subtracted from the main signal, restoring the signal to its original state and reducing the noise introduced between encoding and decoding.

Dolby SR

The late 1980s saw the introduction of Dolby SR – Spectral Recording – which gives greater noise reduction of around 25 dB. It has been successful in helping to prolong the useful life of analogue tape machines, both stereo mastering and multitrack, in the face of the coming of digital tape recorders. Dolby SR differs from Dolby A in that whereas the latter leaves the signal alone until it drops below a certain threshold, the former seeks to maintain full noise reduction (i.e.: maximum signal boost during recording) across the whole frequency spectrum until the incoming signal rises above the threshold level. The band of frequencies where this happens is then subject to appropriately less boost. This is rather like looking at the same process from opposite directions, but the SR system attempts to place a comparably high recording level on the tape across the whole frequency spectrum in order that the dynamic range of the tape is always used optimally.

This is achieved by ten fixed and sliding-band filters with gentle slopes. The fixed-band filters can vary in gain. The sliding-band filters can be adjusted to cover different frequency ranges. It is therefore a fairly complex multiband system, requiring analysis of the incoming signal to determine its energy at various frequencies. Spectral skewing and anti-saturation are also incorporated (see ‘Dolby C’, above). Dolby SR is a particularly inaudible noise reduction system, more tolerant of level mismatches and replay speed changes than previous systems. A simplified ‘S’-type version has been introduced for the cassette medium, and is also used on some semi-professional multitrack recorders.

dbx

dbx is another commonly encountered system. It offers around 30 dB of noise reduction and differs from the various Dolby systems as follows. dbx globally compresses the incoming signal across the whole of the frequency spectrum, and in addition gives pre-emphasis at high frequencies (treble boost). It is not level dependent, and seeks to compress an incoming signal with, say, a 90 dB dynamic range into one with a 60 dB dynamic range which will now fit into the dynamic range capabilities of the analogue tape recorder. On replay, a reciprocal amount of expansion is applied together with treble de-emphasis.

Owing to the two factors of high compansion ratios and treble pre- and de-emphasis, frequency response errors can be considerably exaggerated. Therefore, dbx type 1 is offered which may be used with professional equipment and type 2 is to be used with domestic equipment such as cassette decks where the noise reduction at high frequencies is relaxed somewhat so as not to exaggerate response errors unduly. The degree of compression/expansion is fixed, that is it does not depend on the level of the incoming signal. There is also no division of noise reduction between frequency bands. These factors sometimes produce audible modulation of background hiss with critical programme material such as wide dynamic range classical music, and audible ‘pumping’ noises can sometimes be heard. The system does, however, offer impressive levels of noise reduction, particularly welcome with the cassette medium, and does not require accurate level alignment.

telcom c4

The ANT telcom c4 noise reduction system arrived somewhat later than did Dolby and dbx, in 1978. Capitalising on the experience gained by those two systems, the telcom c4 offers a maximum noise reduction of around 30 dB, is level dependent like Dolby, and also splits the frequency spectrum up into four bands which are then treated separately. The makers claim that the c4 system is less affected by record/replay-level errors than is Dolby A. The system works well in operation, and side-effects are minimal.

There is another system offered by the company, called ‘hi-com’, which is a cheaper, simpler version intended for home studio setups and domestic cassette decks.

Line-up of noise reduction systems

In order to ensure unity gain through the system on recording and replay, with correct tracking of a Dolby decoder, it is important to align the noise reduction signal chain. Many methods are recommended, some more rigorous than others, but in a normal studio operation for everyday alignment, the following process should be satisfactory. It should be done after the tape machine has been aligned (this having been done with the NR unit bypassed).

Figure 7.5   Dolby level is indicated on Dolby units using either a mechanical meter (shown left), or using red and green LEDs (shown right). The meter is normally aligned to the ‘18.5 NAB’ mark or set such that the two green LEDs are on together

For a Dolby A encoder, a 1 kHz tone should be generated from the mixer at +4 dBu (usually PPM 5), and fed to the input of the NR unit. The unit should be in ‘NR out’ mode, and set to ‘record’. The input level of the NR unit should normally be adjusted so that this tone reads on the ‘NAB’ level mark on the meter (see Figure 7.5). The output of the unit should then be adjusted until its electrical level is also +4 dBu. (If the tape machine has meters then the level can be read here, provided that these meters are reliable and the line-up is known.)

It is customary to record a passage of ‘Dolby tone’ (in the case of Dolby A) or Dolby Noise (in the case of Dolby SR) at the beginning of a Dolby-encoded tape, along with the other line-up tones (see ‘Record Alignment’, Chapter 6). During record line-up, the Dolby tone is generated by the Dolby unit itself, and consists of a frequency-modulated 700 Hz tone at the Dolby’s internal line-up reference level, which is easily recognised and distinguished from other line-up tones which may be present on a tape. Once the output level of the record Dolby has been set then the Dolby tone button on the relevant unit should be pressed, and the tone recorded at the start of the tape.

To align the replay Dolby (set to ‘NR out’, ‘replay’ mode), the recorded Dolby tone should be replayed and the input level adjusted so that the tone reads at the NAB mark on the internal meter. The output level should then be adjusted for +4 dBu, or so that the mixer’s meter reads PPM 5 when switched to monitor the tape machine replay.

For operation, the record and replay units should be switched to ‘NR in’.

Dolby SR uses pink noise instead of Dolby tone, to distinguish tapes recorded with this system, and it is useful because it allows for line-up of the replay Dolby in cases where accurate level metering is not available. Since level misalignment will result in response errors the effects will be audible on a band of pink noise. A facility is provided for automatic switching between internally generated pink noise and off-tape noise, allowing the user to adjust replay-level alignment until there appears to be no audible difference between the spectra of the two. In normal circumstances Dolby SR systems should be aligned in a similar way to Dolby A, except that a noise band is recorded on the tape instead of a tone. Most systems use LED meters to indicate the correct level, having four LEDs as shown in Figure 7.5.

Operational considerations

A word may be said about varispeed. It is not uncommon for the replay speed of a tape to need to be adjusted slightly to alter pitch, or total playing time. In creative work massive amounts of speed change are sometimes employed. The pitch change means that Dolby decoding will be inaccurate since the frequency bands will not now correspond to those during the recording process, and Dolby mistracking will result.

Professional noise reduction systems are available as single-channel units, stereo packages, and conveniently grouped multiples of 8, 16 and 24 for multitrack work. They generally fit into standard 19 inch (48 cm) racks. Certain models are designed to fit straight into multitrack recorders so that the complete recorder plus noise reduction combination is conveniently housed in one unit.

Each noise reduction channel is manually switchable between encode for record and decode for replay, and in addition a special input is usually provided which accepts a remote DC signalling voltage, which will switch the unit into encode. Removal of the DC causes the unit to revert to decode ready for replay. Professional tape machines can usually provide this DC requirement, linking it to the record status of each track. Those tracks which are switched to record will now automatically switch the appropriate noise reduction channels to encode ready for recording. The system enables the selection of correct noise reduction status to be left to the recorder itself which is a very convenient feature particularly when a large number of channels are in use.

Single-ended noise reduction

General systems

Several companies offer so-called ‘single-ended’ noise reduction systems, and these are intended to ‘clean up’ an existing noisy recording or signal. They operate by sensing the level of the incoming signal, and as the level falls below a certain threshold the circuit begins to roll off the treble progressively, thereby reducing the level of hiss. The wanted signal, being low in level, in theory suffers less from this treble reduction than would a high-level signal due to the change in response of the ear with level (see Fact File 2.2). High-level signals are left unprocessed. The system is in fact rather similar to the Dolby B decoding process, but of course the proper reciprocal Dolby B encoding is absent. The input level controls of such systems must be carefully adjusted so as to bring in the effect of the treble roll-off at the appropriate threshold for the particular signal being processed so that a suitable compromise can be achieved between degree of hiss reduction and degree of treble loss during quieter passages. Such single-ended systems should be judiciously used – they are not intended to be left permanently in circuit – and value judgements must always be made as to whether the processed signal is in fact an improvement over the unprocessed one.

If a single-ended system is to be used on a stereo programme, units which are capable of being electronically ‘ganged’ must be employed so that exactly the same degree of treble cut is applied to each channel; otherwise varying frequency balance between channels will cause stereo images to wander.

Noise gates

The noise gate can be looked upon as another single-ended noise reduction system. It operates as follows. A threshold control is provided which can be adjusted such that the output of the unit is muted (the gate is ‘closed’) when the signal level falls below the threshold. During periods when signal level is very low (possibly consisting of tape or guitar amplifier noise only) or absent the unit shuts down. A very fast attack time is employed so that the sudden appearance of signal opens up the output without audible clipping of the initial transient. The time lapse before the gate closes, after the signal has dropped below the chosen threshold level, can also be varied. The close threshold is engineered to be lower than the open threshold (known as hysteresis) so that a signal level which is on the borderline does not confuse the unit as to whether it should be open or closed, which would cause ‘gate flapping’.

Such units are useful when, for instance, a noisy electric guitar setup is being recorded. During passages when the guitarist is not playing the output shuts down so that the noise is removed from the mix. They are sometimes also used in a similar manner during multitrack mixdown where they mute outputs of the tape machine during the times when the tape is unmodulated, thus removing the noise contribution from those tracks.

The noise gate is frequently heard in action during noisy satellite link broadcasts and long-distance telephone-line operation. An impressive silence reigns when no-one is talking, but when speech begins a burst of noise abruptly appears and accompanies the speaker until he or she stops talking. This can sometimes be disconcerting for the speaker at the other end of the line because he or she gets the impression that the line has been cut off when the noise abruptly disappears.

Noise gates can also be used as effects in themselves, and the ‘gated snare drum’ is a common effect on pop records. The snare drum is given a heavy degree of gated reverb, and a high threshold level is set on the gate so that around half a second or so after the drum is hit the heavy ‘foggy’ reverb is abruptly cut off. Drum machines can mimic this effect, as can some effects processors.

Digital noise extraction

Extremely sophisticated single-ended computer-based noise reduction systems have been developed. A given noisy recording will normally have a short period somewhere in which only the noise is present without any programme, for instance the run-in groove of an old 78 rpm shellac disc recording provides a sample of that record’s characteristic noise. This noise is analysed by a computer and can subsequently be recognised as an unwanted constituent of the signal, and then extracted electronically from it. Sudden discontinuities in the programme caused by scratches and the like can be recognised as such and removed. The gap is filled by new material which is made to be similar to that which exists either side of the gap. Not all of these processes are currently ‘real time’, and it may take several times longer than the progamme’s duration for the process to be carried out, but as the speed of digital signal processing increases more operations become possible in real time.

Recommended further reading

Dolby, R. (1967) An audio noise reduction system. J. Audio Eng. Soc., vol. 15, pp. 383–388

Dolby, R. (1970) A noise reduction system for consumer tape applications. Presented at the 39th AES Convention. J. Audio Eng. Soc. (Abstracts), vol. 18, p. 704

Dolby, R. (1983) A 20 dB audio noise reduction system for consumer applications. J. Audio Eng. Soc., vol. 31, pp. 98–113

Dolby, R. (1986) The spectral recording process. Presented at the 81st AES Convention. Preprint 2413 (C-6). Audio Engineering Society

See also General further reading at the end of this book.