NICAM stereo: an overview

The diagrams from the original printed report weren't carried over to the web, so a few passages refer to figures that aren't shown.

Modern sound systems are based on digital technology. The first commercial digital sound medium was the compact disc, launched jointly by Philips and Sony in 1982. Pulse-code modulated sound was first introduced in the video field with the Video-8 format. More recently, digital audio tape (DAT) and the soon-to-be-released digital compact cassette are frequently referred to.

Audio signals must start and end in analogue form: what goes into the microphone and comes out of the loudspeaker is a sound pressure wave. Conversion to digital form, by quantisation (sampling the audio signal at regular intervals and assigning each sample a digital value), is done to overcome deficiencies in the intervening systems and equipment, since digital signals are less affected by noise, distortion and so on. When reproduced, the quality of digital sound, and particularly its signal-to-noise ratio, depends largely on the number of quantisation levels used. The ultimate, as used in CD and R-DAT (R for rotary head), is 16 bits, giving 65,536 values, each representing a signal amplitude level. A 16-bit signal would need a wideband channel for transmission, however. As there isn't room for such a signal in the broadcast TV bands, the NICAM system uses 10-bit words. Ordinarily these would give a signal-to-noise ratio of just 60 dB. Complex digital processing improves this to 84 dB, as provided by a 14-bit system.

The 32 kHz sampling rate used by NICAM gives a frequency response of 0–15 kHz in two channels for stereo. Separation between the two channels is total, so dual-language soundtracks can be broadcast without crosstalk. NICAM can carry sound in one or both channels and is completely separate from, and independent of, the existing FM sound channel at 6 MHz. Before it was adopted, the system was tested under many different conditions. It proved extremely rugged, providing high-quality stereo even when the signal is so weak that the picture can hardly be locked.

Technical overview

Many of the processes and techniques NICAM uses are common to those of the compact disc system.

The studio sound feeds for the left and right channels are first pre-emphasised and passed to an analogue-to-digital converter, which produces 14-bit samples (words) at a sampling rate of 32 kHz. These samples go to a compressor that reduces each word to 10 bits, effectively extracting the ten most useful bits and discarding the rest. At this stage an error bit is added for error checking and correction at the receiver. In the following interleaver stage the bits are shuffled like a pack of cards, to spread the effects of propagation errors or interference when the original order is restored at the receiving end. Scrambling is then done by adding an effectively random data stream to the interleaved data, so that the sidebands of the transmitted NICAM signal have no strong, fixed-frequency components that could cause interference.

Modulation comes next. For the UK system the NICAM carrier is at 6.552 MHz. It is phase-modulated by the data pulses using DQPSK. With digital signals, phase modulation is economical of bandwidth, which matters when squeezing another signal into an already tightly packed 8 MHz slot in the TV band. The NICAM carrier is added to the 6 MHz FM sound carrier fed to the transmitter, where it joins the vision carrier (at a level of −20 dB relative to the peak vision carrier) on its way to the transmitting aerial.

At the receiver, the tuner converts the NICAM carrier to an IF of 32.95 MHz. This is demodulated and selected by a filter for the first section of the NICAM decoder, the DQPSK demodulator, from which the NICAM signal emerges as a data stream. It then undergoes a series of processes that reverse those carried out at the broadcasting end. First a descrambler subtracts the random data added at the transmitter to smooth the sidebands. De-interleaving then takes place according to a simple codebook held in memory, restoring the order of the original 10 + 1 bit words. The words are next expanded to 14-bit form according to a scale factor transmitted with the data, and error correction is carried out using the parity bits. At the end of this process we have a real-time 14-bit data stream containing information for both left and right sound channels.

It then remains to convert the data back to analogue form with a D/A converter, and filter the result to recover the original sound signals. User controls are built into the audio amplifiers, whose design is critical if the low distortion and high signal-to-noise ratio are to be maintained. The performance, rating and positioning of the loudspeakers are also crucial to realising the full potential of NICAM sound.

System detail

Pre-emphasis

No sound or vision transmission or recording system is complete without pre-emphasis, which boosts the high-frequency components of the signal. During reproduction the balance is restored by reducing the amplitude of the higher frequencies, and noise is reduced in proportion. The pre-emphasis and de-emphasis characteristics follow CCITT recommendation J.17, with a boost of 6.5 dB at 800 Hz.

Sampling and quantisation

The sampling rate for A/D conversion is 32 kHz, so digitised samples of the original analogue waveform occur at 31.25 µs intervals. To prevent aliasing (which occurs when the lower sidebands of the sampling frequency overlap the upper end of the audio spectrum) and the distortion it causes, the audio must be limited to at most half the sampling rate, 16 kHz. For this purpose a filter with a very sharp cut-off at 15 kHz is inserted in the L and R signal paths, and sampling of the two channels is carried out simultaneously.

Digital compression

The 14-bit words are reduced to 10 bits to conserve bandwidth. The significance of each bit in a word depends very much on the sound level the word represents. The least significant bit can influence the final signal by only about one part in 16,000. That may matter in a delicate, quiet passage, but it matters not one jot when the sample represents a very loud sound or a rapid transition between two widely different levels.

Consider the possible combinations that make up a 14-bit word. The most significant bit (the 14th) passes through the compression system unchanged. The 13th bit is discarded if it is the same as the 14th; the 12th is likewise discarded if it is the same as the 13th and 14th; and so on with the 11th and 10th bits. This leaves words of between 10 and 14 bits. If a word still has more than 10 bits, enough bits are removed from the least significant end to reduce it to 10. The signal has thus been compressed, and as long as the decoder knows what the compressor did, it can expand it back to 14 bits at the receiver.

Protection and scale factor

Each digital word needs some protection against transmission and reception errors. For this a parity bit is added to the end of each word, used in the decoder as a check on the word's six most significant bits. The system uses even parity.

We now have a protected 11-bit word. Next the decoder must be sent a scale factor to indicate which bits were deleted during compression. Bits removed from the most significant end can be re-created from the scale factor; bits removed from the least significant end are lost forever, but that doesn't matter, as those words aren't critical from a noise point of view.

Large signals get 10-bit resolution (coding range 1), coding range 2 gets 11-bit resolution, range 3 gets 12-bit, and so on to the 14-bit range 5, which covers the quietest and most vulnerable sounds. This progressive coding is very effective, saving bandwidth while losing little compared with linear coding.

For scale-factor signalling, the 11-bit (10 + 1 parity) words are grouped into blocks of 32, each lasting 1 ms. With each block a 3-bit code word is sent to indicate the scale factor. There is some inaccuracy in this: since only one scale factor is sent for 32 consecutive words, some words won't receive optimum expansion. What the scale factor actually tells the decoder is the magnitude of the largest sample in each block. The approximation increases quantisation noise in the final audio, but that noise always occurs where the busyness of the signal masks it. Scale-factor data is sent often enough for the fastest perceptible loudness changes to be tracked, and this is the key to subjective noise reduction. It's this running-approximation behaviour that gives the system the first two words of its name, near-instantaneous. The average error is very small in practice and imperceptible to the listener.

So far we've covered two of the fiddle factors in NICAM coding: bit reduction for compression, and one-in-32 scale-factor signalling. Now for the biggest fiddle factor, the way the scale factor is sent. In effect it comes for free, carried in the parity bits of the data words. The scale factor is sent by modifying the parity bits according to a complex table held in memory at both transmitter and receiver. The receiver's decoder extracts the 3-bit scale-factor words using majority-decision logic, a process that also restores the original parity pattern.

There are five different compression patterns, each assigned a coding range from 1 to 5. The table shows how the three-bit code words convey scale factor and protection data.

Scale-factor signalling

Three scale-factor bits give eight combinations. As shown, some of the spares are used to tell the decoder how much error protection each sample needs. The receiver can use this to give extra protection to the most significant bits of the samples.

Interleaving

The parity system gives the data words sufficient protection. But impulsive interference and similar corruption could blow large holes in the data stream, so further measures are needed. The degree of protection required depends on how likely, and in what pattern, the data is to be corrupted. NICAM data coming off air doesn't need the massive error protection used for CD, Video-8 PCM and DAT. Far from the intricacies of CIRC, the interleaving used in NICAM is simple.

It works by writing the data into memory and reading it out non-sequentially. The read-out order, held in a ROM address sequencer, ensures that bits that were originally adjacent are separated by at least fifteen other bits. An error burst in the transmission path can corrupt several consecutive bits in the broadcast stream, but because the error is spread across several words, the damage to each is minor. It can be repaired by parity correction and/or error concealment as needed, techniques used universally in PCM sound systems.

Housekeeping

The digital signal now has enough protection to withstand the rigours of transmission. Housekeeping data is added next so that sections of data can be identified and synchronised, and so the decoding process can be controlled.

Transmitted data is arranged in 728-bit frames. Housekeeping data comes first, rather like the field sync pulse in a video waveform. Each frame lasts 1 ms, so the data rate is 728 kbit/s. There are no gaps between frames.

The frame alignment word (FAW) synchronises and sets up the decoding process in the receiver. It is eight bits long and always 01001110. Next come the five bits C0 to C4, used for decoder control and switching. The frame flag bit C0 is set to 1 for eight successive frames and to 0 for the next eight. This unchanging pattern defines a 16-frame sequence, used to synchronise changes in the type of information being carried. Bits C1, C2 and C3 carry application control information as follows.

C3 stays unchanged throughout these options. C4 is set to 1 when the NICAM channel carries the same sound as the main FM sound carrier, and to 0 otherwise. This allows fall-back switching if the NICAM carrier fails, or when a data transmission is in use. The next data block, AD0 to AD10, is currently unused: it's reserved for additional data, and it remains to be seen what form that takes.

The sound/data block

The initial section of the frame accounts for 24 bits. The remaining 704 carry stereo or dual-channel sound data: sixty-four 11-bit words, with A channel (stereo left) and B channel (stereo right) samples sent alternately, 32 of each. Because two sound channels plus the initial data are transmitted as a single serial stream, the bit rate of each channel is roughly doubled. This form of time compression is a common technique called time-division multiplexing (TDM). It's easy to implement by writing the data into RAM in real time and reading it out at suitable intervals. TDM is used in many data storage and transmission systems to match signal density to channel bandwidth.

This order of A and B samples is the sequence before bit interleaving. Only the 704 bits of sound data are interleaved. For mono sound the data block is arranged slightly differently.

For a mono transmission, two 32-word blocks (n and n + 1) are placed end to end in a single frame, carrying the complete mono signal. Since the data rate is unchanged, the next frame is free for other data if required. Odd-numbered frames carry the mono sound data and even-numbered ones the information data. To indicate this, control bits C1–C3 change to 100, switching the decoder to mono mode.

If two independent mono signals M1 and M2 are transmitted (for example dual-language sound), M1 is carried in the odd-numbered frames as just described and M2 in the even-numbered frames. Control bits C1–C3 change to 010 to switch the decoder's sound routing.

Data can take many forms; computer software is perhaps the most likely. If conditional or restricted access is needed by the broadcaster or their agent, the spare bits C3 and AD0–AD10 may be used to provide it. No data format for this has yet been published.

Scrambling

Having looked at the NICAM data stream in detail, it should be obvious that it contains certain fixed patterns. Since the data modulates a carrier whose sideband structure depends on the modulation, the energy in the sidebands can be evened out by scrambling the data at random, a form of energy dispersal. Truly random scrambling isn't possible, because there would be no way to descramble at the receiver. Instead a simple circuit generates bits in an apparently random but predictable and repeatable way: a pseudo-random sequence generator (PRSG), consisting of a clocked shift register with one or more taps feeding back to its input. The PRSG used for NICAM has nine stages, with a tap between the fifth and sixth feeding one input of an exclusive-OR gate whose other input comes from the final stage. If you work it out, the sequence starts 0000 0111 1011… Once running, the sequence lasts 511 (2⁹ − 1) bits, so it runs through almost one and a half times per frame.

One PRSG is used at each end of the system, transmitter and receiver. Each is reset to 111111111 at the end of the last bit of the frame alignment word (which isn't scrambled), so the PRSGs run in step for the rest of the frame. The PRSG sequence is added modulo-two to the data stream on its way to the modulator; modulo-two addition is exactly what an exclusive-OR gate does.

Modulation and transmission

There are various ways of modulating a carrier with information. The best known are amplitude modulation, used for terrestrial vision broadcasting and radio, and frequency modulation, used for satellite vision, TV sound and FM radio. A digital signal has only two states, 1 and 0, and the most bandwidth-economical way to modulate with it is phase modulation: the carrier frequency stays constant but its phase is altered by the data. The simplest phase modulation is two-phase, with, say, 1s setting the phase to 0° and 0s to 180°, which is easy to detect at the receiver. Bandwidth can be saved by using four-phase modulation, where the carrier has four possible states: 0°, 90°, 180° and 270°. More complex schemes with eight phases exist, but the smaller the phase shift, the harder reliable detection becomes.

The quadrature phase-shift keying used for NICAM has four carrier phase states. At the transmitter, a serial-to-two-bit-parallel converter turns the serial data into a series of bit pairs, each 00, 01, 10 or 11. Each moves the phase by a different amount:

So the carrier has four possible states and moves from one to another depending on each bit pair. With a 00 pair, the carrier stays where it is.

Because each phase change occurs at two-bit intervals rather than once per bit, and the maximum shift is half a carrier cycle either way, DQPSK greatly reduces sideband spread. The sudden phase jumps can be smoothed to reduce the sidebands further. With a bit-pair rate of 364 kbit/s and a carrier of 6.552 MHz, there are at least eighteen carrier cycles between phase changes. Feeding the data to the DQPSK modulator through a low-pass filter changes it from a square wave to something closer to a triangular one, smearing the phase changes out in time so they're spread more evenly across the eighteen available cycles.

With broadcast System I as used in the UK, the normal FM sound carrier is 6 MHz above the vision carrier at a relative level of −7 dB. For use with NICAM it's reduced to −10 dB. The NICAM carrier is at 6.552 MHz (exactly nine times the bit rate) and is transmitted at −20 dB relative to the vision carrier. These levels are quoted as voltages, so the power ratio of peak vision carrier to NICAM carrier is 100:1.

At a main transmitter, the NICAM carrier is fed to the klystron power amplifier used for the FM sound signal. For best efficiency this normally runs in a very non-linear mode and is very sharply tuned to the channel frequency. For NICAM it must be readjusted for more linear operation and greater bandwidth, to accommodate the extra phase-modulated carrier. Great care is taken to keep the NICAM carrier's sidebands low and quiet. There's no interference to the other carriers in the channel, to adjacent sound and vision channels, or, under co-channel conditions, to French System L transmissions, which have an amplitude-modulated sound carrier at 6.5 MHz.

NICAM has been adopted by some European countries and many others have expressed interest. For the continental System B/G, with its FM sound at 5.5 MHz, the NICAM carrier is set at 5.85 MHz. That's rather close to the FM sound carrier but can be accommodated with careful modification of the data-shaping sideband filter, reducing the NICAM signal bandwidth from 700 to 500 kHz.

The NICAM format is compatible with the sound and data arrangements of the MAC/packet system adopted for BSB satellite TV, if that ever gets off the ground again. So chip sets developed for one can be used for the other, which helps both set designers and chip makers. These chip sets have to unpick the rather complex NICAM transmissions at minimal cost, no mean task, and so they are still relatively expensive at around £30 to £40 a set.