The difference between noise and jitter

Noise and jitter in electronics are closely related phenomena. Both are unwanted electrical energy that we mitigate where possible and otherwise work around. Weak signals are obscured by noise. No amount of filtering can reduce noise that is caused by fluctuations intrinsic to the signal being measured, but this noise can be partially eliminated by bandwidth limiting and the signal being investigated can be further purified by waveform averaging. Both operations can be performed by turning a knob or selecting a menu in a digital storage oscilloscope.

The other category of noise is external to the signal of interest. It may be introduced by means of electromagnetic interference, mechanical vibration, acoustical pickup, light pollution or other sources.

In an oscilloscope time-domain display with a sine wave at the input, noise appears as a visible thickening or blurring of the trace, not always prominent. Where there is high noise content, it eventually becomes overwhelming so that the trace occupies a large part of the screen and distorts to the point where triggering is lost.

Noise is also realistically displayed in a spectrum analyzer or oscilloscope in the frequency domain, where noise amplitude is indicated by the height as measured along the Y-axis of spurious signals appearing at unwanted frequencies measured on the X-axis. Then there is the noise floor, which is present in all instrumentation.

Electricians see this as phantom voltage displayed in a multimeter when the probes are not touching energized terminals. It disappears when the probes are touched together. The noise floor also corresponds to snow in an analog TV tuned to a channel that is not receiving a strong transmission and similarly to static in a detuned radio receiver.

The noise floor in a swept-spectrum or real-time spectrum or in an oscilloscope in the FFT mode appears as an irregular roughly horizontal line at a lower amplitude than displayed signals applied to the instrument’s input(s). Real signal information that is at lower amplitude than the noise floor cannot be displayed. Engineers at the major instrumentation manufacturers are continually working to lower noise floors.

Noise, intrinsic (thermal) or otherwise, is a wide-spectrum phenomenon. Therefore, it can be substantially reduced by bandwidth limiting. You can insert an appropriate capacitance in parallel with the circuit upstream from the measuring instrument’s input. This will shunt out high-frequency signal components which constitute noise, but it will not reduce thermal noise originating in the instrument’s internal circuitry. Since the early analog oscilloscope epoch, manufacturers have provided internal circuitry by which the user can temporarily activate one or more levels of internal bandwidth limiting by turning a knob. The trace shown in the display becomes sharper and triggering, if it was suppressed, is restored. The only problem with bandwidth limiting is that the signal under investigation may also have a frequency spectrum that extends beyond the imposed bandwidth limit. If that is the case, the signal will be discarded along with the noise.

To demonstrate, in a Tektronix MDO3000 Series Oscilloscope we feed a sine wave from the internal arbitrary function generator into the Channel One analog input, and display it on the screen. Pressing Output Settings, we add 30% noise. That is enough to blur the trace and distort the waveform to the extent that it loses triggering. Cycling the Channel On button, the horizontal display menu appears and we press the soft key associated with bandwidth. The vertical bandwidth menu comes up, letting the user choose the level of bandwidth limiting, 1 GHz (full), 250 MHz, or 20 MHz. 250 MHz reduces the noise a slight amount, but for this particular AFG signal with 30% noise added, triggering remains impaired, so we have to turn the bandwidth down to 20 MHz. The signal is considerably less noisy and triggering is restored.

Waveform averaging is more effective than bandwidth limiting and there is not the problem of rejecting a high-bandwidth signal. To see how it works, turn bandwidth limiting back to Full. Then, with the same sine wave again with 30% noise displayed, press Acquire>Mode. The vertical acquisition menu appears. Pressing the soft key associated with Average, we see that Multipurpose Knob a can select the number of waveforms that are averaged. When the minimum number, 2, is selected, the sinewave remains very distorted and unstable. At 128 samples, the waveform is still distorted, but triggering has been restored.

Scrolling the display up through the logarithmic series of waveforms to be averaged — at the maximum, 512 — the trace is sharp, indicating noise has been eliminated. The trace now appears to pulsate. That is because a certain finite amount of time is required for the oscilloscope to average and display the 512 discrete events.

Waveform averaging is highly effective in eliminating noise, which is entirely random. The waveforms do not vary or if they do the variance is quite slow compared to the noise and the rate of samples taken to be averaged. As they are averaged, the successive waveforms reinforce while the random noise cancels out.

As for jitter, like noise, it is an undesired distortion of the signal. Jitter, however, happens in the digital realm and, rather than an unwanted variation in amplitude of an analog signal, it pertains to the timing of digital pulses. Of particular interest are the timing of the zero crossings and the timing of digital output transitions, which in turn depend on period signal crossings of decision thresholds that determine whether a bit is 1 or 0. In all cases, when there is jitter, the data stream as conveyed to the receiver will contain inaccurate information.

Like noise in general, jitter is often caused by electromagnetic interference or crosstalk. Both of these can be mitigated by identifying the sources and powering them down, or by creating greater spatial separation by rerouting wiring, by relocating equipment, or by placing shielding between source and affected data cable.

Jitter affects computer monitors, causing them to flicker, can degrade the operation of processors in computers and test instrumentation, distorts audio signals and garbles data in networks. Where possible it should be suppressed, except perhaps when used for special effects in music synthesis.

The principle types of jitter are:
• Absolute jitter, which is a measure of the deviation in time of a clock pulse edge from its ideal location.
• Period jitter, which is a variation between the ideal clock periods and actual clock periods. Synchronous circuitry as in a central processing unit is seriously impacted by these variations.
• Inter-cycle jitter, which is the difference in duration between successive clock periods. When this is excessive, microprocessors cannot function.

Jitter often but not always is characterized by a Gaussian distribution. It may be non-Gaussian when caused by a power supply or other external noise. Jitter, unfortunately, has a large presence in computer networking, where it appears as packet delay variations. Due to the disruption in timing, packets are lost, a serious problem in voice-over-IP. Service may be improved by implementing effective buffering at the receiver.

Another way to categorize jitter is to distinguish between random jitter and deterministic jitter. Because it is caused by omnipresent thermal noise in the electronic circuitry, random jitter conforms to a Gaussian distribution. That is due to the Central Limit Theorem, which states that the composite effect of many uncorrelated noise sources, without regard to their individual distributions, approaches a normal distribution.

Deterministic jitter differs from random jitter in that it is constant and predictable. Its peak-to-peak value is bounded and its distribution is non-Gaussian. Examples are duty-cycle dependent jitter and inter-symbol interference. These two varieties constitute total jitter.