Real-World Technology Ltd

FM Threshold: "What -- 8dB? I can do better than that!"

The Importance of Threshold

Perusing the ads, we might think that today's satellite receivers are no good unless they boast a carrier-to-noise threshold of 6dB, or maybe 4dB, or even better. After all, how technology has advanced since the start of low-cost TVRO in 1979! (or Astra DTH in 1989 -- everyone has their own "Year Dot.") But in the real world, what do these claims really mean?

Above FM threshold, the recovered baseband signal to noise ratio (S/N) varies according to the change in pre-demodulation carrier to noise ratio (C/N or CNR) in a linear manner. S/N is always greater than C/N, more so for wider deviation signals, but for a given signal a 1dB reduction in C/N causes a 1dB fall in S/N.

The commonly agreed definition of FM demodulator threshold is the value of C/N (as C/N is reduced) where this dB-for-dB relationship fails, by a factor of 1dB. That is, the recovered S/N value is 1dB lower than would be predicted by extending the straight line S/N vs C/N graph to lower values of C/N:
This "knee" in the characteristic is due to "threshold effects", which usually means large amplitude random bipolar impulse noise spikes. The character of the video noise changes: above threshold it has a smooth and even granularity, with the triangular spectrum of FM as modified by de-emphasis; the additional noise as we reach threshold is impulsive and distracting. We have a case of what US CATV operators of the 1970s dubbed "the sparklies".

And because the graph steepens below the knee (to an extent which depends upon video deviation), every dB loss of signal results in many more sparklies, with the picture rapidly becoming unwatchable.

What C/N value can we use?

Now what matters to the viewer is a clean picture. Operation at threshold is not acceptable, for by then the dancing black and white spots are already spoiling the picture. The viewer demands operation at or above what we at RWT call the "sparklie extinction" C/N, which is always higher than the "threshold" C/N.

But increasing C/N is expensive. For a given power flux density (satellite signal level on the ground) we can only deliver a higher C/N by upgrading the dish and (to a lesser extent) the LNB. So achieving the lowest practical threshold in the receiver is important.

Even if we could lower the threshold without limit, we would eventually reach a point where the (extended) linear relationship of S/N to C/N made the picture unwatchable, not through the spiky noise of threshold but purely from linear video noise -- as with terrestrial AM TV reception in a weak signal area, the picture would dissolve into snow.

This practical lower limit for C/N under extended threshold conditions depends upon the video deviation of the transmission. With a PAL or NTSC signal a 33 or 36 MHz transponder might operate at 21 to 25 MHz/V deviation, but in a 26 MHz transponder the maximum possible deviation is 16 MHz/V, and an 18 MHz half-transponder carrier might only have 9 MHz/V (that's peak to peak) deviation.

Taking the 16 MHz/V (e.g. Astra) case, a normal operating C/N of 14dB delivers a weighted PAL video S/N of 45dB -- between "good" and "excellent" on the CCIR 5-point scale. Assume a threshold extension system permitted operation at a 3dB C/N. Linear video noise would by then have degraded S/N by 11dB, to a weighted value of 34dB -- just on the low end of the "fair" range.

Wider deviations fare better: with a full 36 MHz transponder, pictures would be classed "fair" to "good" even at a threshold-extended C/N of 3dB.

But what threshold value can be achieved, and how?

A "straight" discriminator (FM detector) exhibits a C/N threshold of around 10 or 10.5 dB. Sparklie extinction occurs at about 12.5 to 13 dB C/N, so this is the minimum value of C/N for a clean picture. In order to improve on this, the threshold extension demodulator was developed. This employs a phase-locked loop (PLL) demodulator, or functionally equivalent circuits such the injection-locked oscillator, dynamic tracking filter or FM feedback, to dynamically narrow the IF or demodulation bandwidth according to modulation (signal content), to make the best of lower C/N ratios.

Narrowing the channel bandwidth reduces the in-channel noise, but we can't simply use a narrower IF filter to extend threshold. Frequency modulation generates a theoretically infinite spectrum of diminishing sidebands, and if we cut off this spectrum too close to carrier frequency we generate distortion products and an effect known as truncation noise.

The higher baseband frequencies generate the most dispersed sidebands, more so with the pre-emphasis used on FM TV, and the energy in these sidebands is greatest where the picture contains sharp edges or transient detail, and (with PAL or NTSC) saturated colours. Truncation of these sidebands manifests itself as sparkly noise on fine detail, caption edges, or areas of strong colour. This truncation noise is similar in quality to the "sparklies" seen at or below threshold, so reducing the static channel bandwidth below a critical value has the opposite effect to that desired -- it worsens the picture.

The optimum IF bandwidth for a PAL TV service with normal subcarrier loading and 16 MHz/V video deviation in a 26 MHz (nominal) channel is around 24 MHz. For a 25 MHz/V transmission in a 36 MHz channel, 32 MHz bandwidth is about right.

The threshold-extension demodulator must then be a compromise to recover the baseband signal from this fixed IF bandwidth, whether it be a dark scene in a black and white movie, or a studio graphic with bright colours and sharp-edged captions.

What is all this "static" versus "dynamic" threshold thing anyway?

A PLL demodulator has parameters of loop gain (below limiting) and bandwidth, which affect its tracking range. Set the gain and bandwidth low and the loop will hang on to an undeviated carrier without visible cycle slips down to what, in the IF bandwidth, is a very low C/N -- 0dB or less. This is because the loop sees the C/N in its own bandwidth, which could be arbitrarily low. But apply an FM TV signal and that same loop will simply fail to track -- only a greyish pattern emerges on the TV screen. The loop parameters must be set to handle the worst case (most difficult) signal normally encountered.

We at RWT take this to be equivalent to a test signal containing 100/0/75/0 colour bars, 60% amplitude multiburst to 5.25 MHz, and 1T pulse and bar, together with one primary and six companded (Wegener-style) subcarriers. A conventional PLL demodulator will track this without sparklies down to about 11.5 dB C/N, which is equivalent to a threshold of 8.5 to 9 dB, using the normal measurement conventions. This is known as its dynamic threshold.

With the same PLL settings and an unmodulated carrier, sparklies start to show at about 9 dB C/N, and we measure static threshold in the region of 7dB.

This is the level of performance achieved by all the major tuners today, excellent products all, be they labelled Sharp, Hitachi, Mitsumi, Alps, Philips, Pace, Telia, Salcomp (etc.), and whatever the claims of the receiver manufacturers who incorporate them.

Then why do people quote the static threshold?

-- Because it makes their product appear to be better than it is! Sure, "7dB static threshold" (or even just "7dB threshold") sounds a whole lot better than "sparklie extinction at 11dB", but it's a clean picture all the time that the customer wants, not just when the screen fades to black. And even quoting dynamic threshold is not perfect, as so much depends on how it is measured -- not all video S/N measurement methods react the same to impulse noise, nor can they all measure it in the presence of a high amplitude 3.58, 4.43 or 5 MHz sine wave. So inevitably some suppliers make claims which are theoretically impossible.

What about the so-called "low threshold" or programmable threshold receivers?

Some of these claim a 4dB or similar threshold value, and the degree of "threshold extension" may be programmed in (typically) 16 steps. What they do is to alter the demodulator PLL parameters in order to trade off dynamic threshold against static. This enables the tuner to make the best of a below-threshold signal, by accepting a degree of truncation distortion (tearing on sharp edges or captions) in exchange for fewer sparklies in "flat" areas of the picture. In this way a picture of sorts can be viewed where at full bandwidth and tracking this would have been impossible. The trick is similar to the use of a too-narrow IF filter, or variable-bandwidth filter, but it is more flexible and less costly to implement. But invariably the threshold claimed is the static value (dynamic is impossible to assess in the presence of truncation noise), and is referred to the noise bandwidth of the original channel. As with the use of a narrow filter to limit the noise bandwidth, the method cannot correctly be described as threshold extension.

So how can true dynamic threshold be further improved?

There are a number of techniques which do extend dynamic threshold, and so the sparklie extinction point, below the normal PLL value. The problem is that most are comparatively expensive to implement, and so cannot be justified in consumer products. One method improves demodulator tracking by employing line and field delays in the PLL error signal, but is not applicable where scrambling systems destroy the correlation between adjacent lines and/or fields. Another uses techniques similar to VCR drop-out compensation, to cancel the effects of sparklies in the video signal. Yet another employs digital signal processing to recognise the signature of a "sparklie" and subtract each occurrence from the signal.

RWT tuners for consumer applications use a simple but effective technique to modify the loop error signal frequency response according to the type of signal being received (PAL, NTSC, MAC etc.). The result is to remove the dependence of dynamic threshold on saturated colours -- the too-familiar effect of bright reds breaking into showers of spots doesn't occur. In this way the true dynamic threshold, and hence the sparklie-extinction point, are extended by 0.5 to 1dB beyond the standard PLL value of most tuners, without degradation of the static threshold. Typically we achieve sparklie extinction at a C/N of 10.5 to 10.75 dB for a full-transponder transmission carrying the test signal described above, which translates to a dynamic threshold of 8 dB.

More specification topics to be added, soon ...
Copyright © 1996 Real-World Technology Ltd.

Specification features of RWT compact tuners.
Existing tuner designs.
The RWT story so far.
Dish Size Chartsfor system planning.
News from RWT.
Links to interesting (!) Web sites.

Return to RWT Home Page