Digital modes like FT8 work below Noise? – A myth
Barring a few modulation modes like the Spread Spectrum, including both direct Sequence Spread Spectrum (DSSS) and the Frequency Hopping Spread Spectrum (FHSS), typically most other commonly used digital modes need signals to be above the noise floor with a positive magnitude of Signal-to-Noise Ratio (SNR). Spread Spectrum due to its nature, has a very low signal density that typically lies below the noise floor. Hence, it is difficult to detect and equally difficult to decode. This is the reason why authorities around the world are wary of allowing radio amateurs to operate Spread Spectrum… But that’s a story for another day.
A typical Communication Receiver – Noise vs Signal
A typical communication receiver or a transceiver hat we use for amateur radio is designed to cater to SSB radiotelephony, CW, and various other forms of digital communication including analog and digital SSTV. The primary filter bandwidth of the receiver chain is typically designed to process signals with up to 2500-2700 Hz baseband channel bandwidth. This filter bandwidth also determines the receiver bandwidth or what is called the Detection Bandwidth in technical terminology. Of course, we also have options to switch in narrower bandwidth filters to cater for better CW experience. However, even with the CW filters switched in the channel bandwidth is usually no less than 250-500 Hz. Even this is far too wide for optimum narrow-band digital text modes.
When tuning across the band to locate signals, it is a usual practice among radio amateurs to select the SSB filter (~ 2500 Hz). This allows listening to various types of signal modulations with ease. Under these circumstances, various narrow-band digital mode transmissions like PSK, MFSK, JT9, JT65, FT8, WSPR, etc may not be audible or not be visible on the band-scope or waterfall displays of the transceiver. Yet, these signals can be copied and perfectly decoded by various software utilities designed to decode these digital modes.
Why is it so? If the digitally modulated transmission is inaudible and below the noise floor, is it not logical to conclude that these didi-modes work well with negative SNR? The answer is an emphatic NO… This brings us to the concepts of Noise bandwidth and Modulation bandwidth. Let us see what they are.
The concepts of Modulation and Noise Bandwidth
Let us first briefly figure out the Modulation Bandwidth of a modulation mode and thereafter correlate it to the Noise bandwidth of the communication channel. This will provide us an insight into what is really going on. We have a separate article on this subject which goes into the details of noise bandwidths under the section on Noise in Radio Communication that you could read later.
Modulation BandwidthDifferent types of radio signal modulations have different modulation bandwidths to carry information. For instance, a typical SSB modulation bandwidth for radiotelephony is around 2500 Hz, while normal CW may require 100-250 Hz bandwidth depending on the keying speed. On the other hand, digital text modes like PSK31 (BPSK31) at 31.25 bauds (~50 WPM) would theoretically need 63 Hz bandwidth, while JT9, FT8 or JT65 operating at very low baud rates need as low as 1.5-6 Hz only. Slower the information transfer rate (Data rate), the lower is the required modulation bandwidth.
Noise Bandwidth (Detection Bandwidth)Intuitively, we all know that the background radio hash noise that we hear on HF radio determines our ability to copy a DX station. Greater, the background noise, the more difficult it is to copy a weak station. We also know that when we switch in a narrower receiver filter, the intelligibility becomes better.
The quantum of aggregate noise on any radio receiver frequency is directly proportional to the width of the filter. The wider the filter bandwidth (detection bandwidth), the greater is the aggregate noise. Every time we double this bandwidth, the noise floor power level in the receiver doubles and vice-a-versa. Hence, the signal level being constant, each time we double the receiver detection bandwidth, we increase the noise power in the receiver chain by +3 dB and consequently degrade the SNR by -3 dB, and so on.
This brings us to a vital question… How do we reduce this background noise? The answer is to use narrower filters to ensure narrower detection bandwidths. How narrow a filter bandwidth can we use? This where we come to the point where we would say that for optimum communication, we need the detection (filter) bandwidth of the receiver to correspond to the modulation bandwidth of the digital mode that we are using. Unnecessarily wide detection bandwidth would result in degradation of SNR which might make the signal not only inaudible but also produce a negative SNR. A perfectly copyable narrow-band modulated digital signal which actually has a healthy positive SNR when copied on a receiver with an optimally narrow detection bandwidth, will appear to have a negative SNR on a typical amateur radio transceiver.
Why do they say that FT8 or JT65 can copy at -26 to -30 dB SNR?
Actually, they don’t… The way it is put across may be quite deceptive and may appear to be misleading to an average amateur radio operator. As I mentioned earlier, all these seemingly magical modes too require a healthy and positive SNR to be able to decode, just like any other modulation including SSB radiotelephony.
This is where we will now revisit the concepts of the receiver bandwidth, filter bandwidth, or detection bandwidth, whatever we might like to call it.
The confusion is because all the negative values of SNR that we find associated with these apparently magical digital modes are all referenced to a typical transceiver setup for SSB with a reference detection (noise) bandwidth of 2500 Hz. If we were to translate each of these negative SNR values associated with such modes to the equivalent required detection bandwidths corresponding to those modes, all the apparent magic will disappear in thin air… Let us unravel the mystery with a typical example below.
OK, if what you say is true, then how come we manage to decode these signals when our transceivers have wide detection bandwidth and drown the signal in noise?… Fair question!!!
The answer is that our transceivers are generally incapable of directly decoding such signals at all without the assistance from outboard narrow bandwidth signal processing systems or software. The output of the transceiver audio that contains noise across the 2500 Hz bandwidth plus the signal that is sunk well below the noise level is fed to external hardware or software-based decoder. Most often we use software decoders running on our personal computers that are connected to the transceiver.
The wideband noise plus signal that is inputted to the decoder is made to pass through a narrow band filter with a noise bandwidth of 2.5 Hz. The signal that is 2.5 Hz wide stays intact while the excess noise around it is now filtered out. As a result of this, the noise magnitude at the output of the software DSP based filter reduces by 30 dB (Log(2500/2.5)). Hence, we succeed in reclaiming the original SNR of +6 dB after this elaborate software-based filtering process. Thereafter, the rest of the software decodes the +6 dB SNR signal to complete communication.
The bottom line is that these digital modes do not work magically with negative SNR. The actual positive SNR is cloaked by the way we usually specify it with reference to 2500 Hz bandwidth. The method of specifying SNR of various modes against a 2500 Hz detection bandwidth reference is practical and convenient for the scientific community and all others who are technically minded. It allows easier assessment of relative performance characteristics of various narrowband digital communication modes. However, among a large section of the amateur radio community, it spreads the shenanigan that these digital modes have magical properties of working effectively with signals far below the noise floor.