What is the "cliff effect" on digital radios, and why do they cut out suddenly instead of just getting noisy?

The cliff effect is the abrupt loss of a digital radio signal at the edge of coverage, with no warning stage in between. A DMR radio sounds clean right up to the point where the error correction can no longer reconstruct the voice frames, and then it stops - no static, no fade, nothing. Analogue radio degrades over roughly 10 to 20 dB of falling signal, getting noisier the whole way; a digital radio does the same transition in about 2 to 3 dB. That is not a fault, it is how forward error correction behaves, and it changes how an in-building system has to be designed and tested.

Why the transition is so sharp

An analogue FM radio sends your voice as a continuously varying carrier. When signal falls, noise is added to the audio and the result is progressively hissier, but a listener can still pull words out of it well below the datasheet sensitivity figure. There is no threshold - just a long slope.

DMR does something different. Voice is sampled, compressed by a vocoder into digital frames, and wrapped in forward error correction before transmission. The receiver's FEC repairs corrupted bits invisibly, so audio quality stays flat while the bit error rate climbs from near zero through one, two, three per cent. Nothing in the audio tells the user this is happening. Once errors exceed what the FEC can correct, whole voice frames fail to decode, the vocoder has nothing to synthesise from, and output collapses within a decibel or two. Graph audio quality against signal level and the digital line falls off a cliff where the analogue line slopes.

The reference point manufacturers publish is 5% bit error rate, and receiver sensitivity for a DMR portable is normally quoted at that threshold rather than at 12 dB SINAD. Beyond 5% BER vocoder audio degrades very quickly. Datasheets for the same class of radio commonly quote a carrier-to-noise requirement in the region of 14 dB at that point - verify against the specific radio's datasheet rather than designing to a remembered figure.

The numbers that matter when designing around it:

• Analogue usable-to-unusable transition: roughly 10 to 20 dB of falling signal. Digital: roughly 2 to 3 dB.

• DMR static sensitivity is typically quoted at 5% BER, around -116 to -119 dBm depending on the model - check the datasheet.

• Design threshold for reliable in-building DMR voice: -95 dBm at the portable, which sits over 20 dB above static sensitivity deliberately.

• Fade margin above that threshold: 10 dB minimum for commercial systems, 15 dB for mission-critical and public-safety use.

• Body loss for a radio worn at the hip: 3 dB, taken off before any of the above.

Why this makes building coverage harder, not easier

On an analogue system, users find your dead zones for you. Somebody walks into a stairwell, hears the hiss rise, and steps back out. The complaint that reaches facilities is "it gets scratchy near the service lift" - a location and a direction. On a digital system the same user hears perfect audio, turns a corner, and is simply gone. They do not know whether they are one metre or twenty metres outside coverage, and neither does anyone else.

That is why the design threshold for a digital in-building system is set at -95 dBm rather than anywhere near the receiver's -116 dBm sensitivity. The gap is not waste. It absorbs body loss, multipath fading as a person moves through a corridor, the difference between a radio held at the head and one clipped at the hip, and the fact that indoor noise floors are rarely as quiet as the bench measurement the sensitivity figure came from. A system designed with 5 dB of headroom on paper will have users falling off the cliff daily.

There is a second consequence that catches people out. The cliff is set by carrier-to-noise, not by absolute signal level. If LED drivers or variable-frequency drives raise the receive noise floor by 10 to 20 dB, the cliff moves inward even though every RSSI reading on the survey is unchanged. A building can pass a signal-level survey in the morning and fail on audio in the afternoon when the plant is running. This is exactly the gap covered in our comparison of DAQ 3.0 against the -95 dBm signal-level criterion, and it is the reason a level-only acceptance test is not sufficient for a digital system.

Worth being clear about what digital does not change: the cliff effect is not a reason to avoid DMR, and switching back to analogue does not improve coverage. Within the covered area digital audio is better, and the covered area itself is broadly similar. We set this out in whether switching to digital DMR radios fixes a building's dead zones. What changes is the tolerance for a marginal design - analogue forgives one, digital does not.

Related questions

Does the cliff effect mean a digital system needs a larger fade margin than an analogue one? In practice yes, though the published margins are the same. With analogue, a 5 dB shortfall produces noisy but usable audio; with digital it produces silence. Design to 10 dB above -95 dBm for commercial systems and 15 dB where the radio is relied on in an emergency.

Can a coverage survey pass and still leave cliff-effect dead spots? Yes, if it measures signal level only. Add a bit error rate or delivered audio quality check at the same test points, and take readings with the building's plant running rather than at night.

Does the cliff move if the building's noise floor rises? It does. The threshold is a carrier-to-noise ratio, so an elevated noise floor from LED lighting, VFDs or a new tenant fit-out pushes the coverage boundary inward with no change in measured signal strength.

How Suneast approaches this

We design in-building two-way radio systems to a delivered level of -95 dBm with an explicit fade margin stated in the report, not to the radio's sensitivity figure, precisely because a digital system gives users no warning before it drops them. Acceptance testing records both signal level and delivered audio quality at every grid point, with the noise floor measured separately so a marginal result can be attributed to a coverage shortfall or to interference rather than guessed at. Where a client's existing analogue system is being migrated to DMR, we re-survey before the swap - an area that was tolerably scratchy on analogue is often below the digital threshold, and that is far cheaper to find before the radios are bought than after.