Below the Noise Floor — Episode 11: "Noise Reduction and AGC"

Noise reduction and AGC are two of the most misunderstood controls in SDR software - not because they are complicated but because they solve different problems. This episode explains what each one actually does, when to use it, when to leave it alone, and how the noise blanker fits into the picture.

Transcript

SPEAKER_00 0:00

Last episode covered Receive Filters, the frequency selective tools that define which parts of the spectrum reach your ears. This episode covers two more receive processing tools that work differently and solve different problems: noise reduction and AGC, automatic gain control. These two controls are often misunderstood, partly because they both improve received audio under some conditions, and partly because the improvement each one provides feels similar on the surface, things sound cleaner, the signals easier to copy, but they are doing fundamentally different things, and using the wrong one for the wrong conditions either produces no improvement or actively makes things worse. Let me take them one at a time. Automatic gain control is the older and simpler of the two concepts. The problem it solves is this. HF signals vary enormously in strength. A station 50 miles away on 40 meters might arrive at your receiver at an extremely strong level. A station 5,000 miles away on the same band at the same moment might be 20 or 30 decibels weaker. If your receiver had a fixed amplification level, it would either amplify a weak signal so much that strong ones overloaded the audio or it would set amplification low enough for strong signals and make weak ones barely audible. AGC automatically adjusts the receiver's amplification to compensate for signal strength variations, keeping the audio output at a relatively consistent level regardless of how strong or weak the incoming signal is. In practical terms, AGC is what makes it possible to hear a mix of strong and weak signals in the same listening session without constantly adjusting the volume. A station calling CQ from nearby comes through at a comfortable level, a DX station from the other side of the planet, much weaker, also comes through at roughly the same level rather than being barely audible. The AGC is continuously measuring signal strength and adjusting gain to compensate. The key parameter of AGC is its time constant, how quickly it responds to changes in signal strength. Either SDR offers several AGC speed settings, typically labeled slow, medium, fast, and sometimes off. FastAGC responds quickly to signal strength changes. It is good for conditions where signals vary rapidly, a heavily fading path, a weak DX signal that fluctuates significantly from moment to moment. Fast AGC tracks the fading and tries to keep the audio level consistent even as the signal swings up and down. The downside of FastAGC is that it can pump. If a brief loud noise or a strong signal appears, the AGC pulls down the gain rapidly and then has to recover, causing a momentary drop in audio level that sounds like a pumping or breathing artifact in the background. Slow AGC responds more gradually, it is better for conditions where signals are relatively stable, a strong regional contact on 40 meters at night, a clear 20-meter contact with minimal fading. Slow AGC provides a more natural audio experience because it does not react to brief fluctuations. The trade-off is that if a very strong signal suddenly appears, a nearby station calls CQ right next to your frequency. Slow AGC takes longer to pull down the gain, and for a brief moment, the audio may be uncomfortably loud before the AGC catches up. Medium AGC is a compromise that works acceptably for most general operating conditions. It is a reasonable default if you're not sure which setting to use. AGC off is occasionally appropriate but requires care. With AGC disabled, the receiver operates at a fixed gain level. This is useful for certain measurement or monitoring tasks, and some operators prefer it for CW operating where they have a good handle on expected signal levels and want the most linear audio response possible. With AGC off on a noisy band, a strong adjacent signal can produce audio that is genuinely uncomfortably loud. Do not disable AGC and then scan across the band without being ready for that. One more thing about AGC worth knowing AGC affects the entire signal coming into the receiver. It does not distinguish between the signal you want and the noise or interference you do not want. If you are listening to a weak DX signal and a very strong adjacent station is just outside your filter, the strong station may be influencing your AGC even though it is outside the pass band. This is called AGC desensitization, and it is one of the reasons filter width and AGC setting are related. A wide filter on a crowded band lets strong adjacent signals interact with the AGC in ways that can reduce the apparent strength of weaker signals you are trying to copy. Narrowing the filter to exclude the strong adjacent station can actually improve the weak signal you want, partly through the filtering itself and partly by letting the AGC respond correctly to just the signal you care about. Now noise reduction. This is a fundamentally different tool. Noise reduction in Ether SDR is a digital signal processing algorithm, specifically a variant of what is called spectral subtraction or wiener filtering, depending on the implementation. It works by analyzing the audio signal in real time, attempting to distinguish between the signal component and the noise component, and selectively reducing the noise component while leaving the signal intact. The way most noise reduction algorithms do this is by building a model of what the noise looks like, its spectral shape, its statistical character, and then subtracting an estimate of that noise from the total signal. The assumption is that noise has a relatively consistent character across time and frequency, while a voice signal or CW signal has a different, more structured character. By comparing the current audio to the noise model, the algorithm can suppress portions of the audio that look like noise and past portions that look like signal. When noise reduction works well, the effect is striking. A voice that was almost buried in atmospheric noise becomes more intelligible. The constant background hiss or crackle of a noisy band reduces noticeably. Weak signals that were at the edge of copyability become clearly readable. When noise reduction works poorly, and this is the part most documentation skips, it produces artifacts. Musical noise is the most common artifact of aggressive noise reduction, a kind of warbling, burbling, electronic sounding background that appears when the algorithm is suppressing too aggressively and incorrectly classifying parts of the signal as noise. If you have ever heard a voice contact where the background sounded oddly melodic or the voice had a strange electronic warble to it, you were probably hearing musical noise artifacts from noise reduction set too high. The level of noise reduction aggressiveness is controlled in Ether SDR by a slider or numerical setting, typically ranging from zero to some maximum value. The correct setting is not the maximum. The correct setting is the lowest level that produces a noticeable improvement without introducing audible artifacts. In practice, this means starting at zero and gradually increasing the noise reduction level while listening. At some point, you will notice the background noise decreasing. Keep increasing until you start to hear artifacts, any warbling, any strangeness in the background quality, any degradation of the target signal's naturalness. Then back off slightly from that point. That is your practical ceiling for current conditions. The practical ceiling varies with conditions. On a very noisy band with strong atmospheric noise, a higher noise reduction setting may be usable before artifacts appear. On a quieter band where the signal is already clean, even a low noise reduction setting may introduce artifacts because there is not much noise for the algorithm to work with and it starts misclassifying signal as noise. On a clean band with a strong signal, noise reduction may produce no improvement at all and should simply be left at zero. One important thing to know: noise reduction on HF works best on voice signals and is essentially useless or counterproductive for digital modes like FT8. FT8 tones have a very specific spectral character that noise reduction algorithms often classify as noise and attempt to suppress. If you are decoding FT8 and WSJTX and have noise reduction active in Ether SDR, turn it off. The FT8 decoder in WSJTX does its own noise processing that is specifically optimized for the FT8 signal structure, and it works better on clean audio than on audio that has already been processed by Ether SDR's noise reduction. Now the noise blanker, which is the third tool in this set and the most specific, the noise blanker, NB in Ether SDR's interface, is designed for a specific type of interference. Brief, high-energy impulse noise, power line arcing, lightning, ignition noise from a nearby vehicle. These sources produce very short, very intense spikes in the audio that are completely different in character from continuous background noise. They are too brief for noise reduction to handle effectively and too intense for AGC to suppress without affecting the surrounding audio. The noise blanker works by setting a threshold. Any audio spike that exceeds that threshold by a significant amount in a very short time window is identified as an impulse and blanked, muted, for the duration of the spike. The effect is to replace the impulse with a brief moment of silence rather than letting the spike through to your ears or downstream processing. On a band plagued by power line noise or near a road with heavy diesel traffic, the noise blanker can make a dramatic difference, turning a constant crackling cacophony into something you can actually copy through. The threshold setting on the noise blanker is critical. Set it too low, and the blanker triggers on legitimate signal peaks as well as noise, blanking parts of the audio you want to hear. Set it too high, and it does not trigger on the interference spikes you are trying to suppress. The right setting is high enough to pass normal signal peaks and low enough to catch the interference. This requires some experimentation on whatever interference you are dealing with. Ether SDR also has a noise blanker 2 setting, sometimes called NB2, which uses a different algorithm, typically a time domain blanker, as opposed to the frequency domain approach of NB. The two blankers behave differently and one may work better than the other depending on the specific character of your interference. If NB is not helping, try NB2 before concluding the blanker cannot help. To summarize the three tools and when to reach for each one, AGC is always on in some form for general receiving. It is the baseline gain management that makes variable strength signals manageable. Use slow AGC for stable conditions, fast for heavily fading paths. Noise reduction is for continuous background noise like atmospheric noise or broadband interference. Start low and increase until artifacts appear, then back off. Leave it off for digital modes. The noise blanker is specifically for impulse noise, power lines, lightning, ignition, and requires threshold calibration to work correctly. None of these tools is a substitute for a good antenna or a quiet operating location, but used correctly, they meaningfully extend what you can copy under difficult conditions. Next episode is the one we have been building toward in the series the Ether SDR voice chain. Episode 12 covers the transmit audio path from microphone to transmitted signal, one control at a time, starting from zero and working toward a clean functional voice signal. This is below the noise floor.