In a previous post we described how phase noise information can be extracted from real-time oscilloscope waveform acquisitions using two different techniques to demodulate the phase. In this article we’ll take a look at the potential accuracy of the serial data clock recovery technique, what kinds of signals can reasonably be analyzed, and some ways to improve such measurements.
In order to check that the oscilloscope phase noise measurement is accurate we can use a clean signal source with a broadband random phase modulation source built-in. By injecting a relatively large PM amplitude over broad frequency we can verify the noise level by comparison with a measurement made on a Signal Source Analyzer such as the Keysight E5052B.
In the measurement below (Fig. 1), the SSA result is in blue and the oscilloscope measurement (using a Keysight MSOS804A) result is in green. There is excellent agreement over the range of injected PM. Above 2 MHz the SSA’s lower noise floor is the reason for the separation in the curves.
Measurement Noise Floor
The measurement floor of a jitter measurement on a real-time sampling oscilloscope is affected by both vertical (voltage) accuracy and timing accuracy. Vertical noise in the sampling system, stability of the timebase, the phase noise of the oscilloscope’s own oscillator & imperfections in the interleaving architecture of the scope will all contribute to errors in the jitter measurements and thus the measured phase noise.
An example of an oscilloscope jitter measurement floor specification is:
The Intrinsic Jitter portion is dependent on the stability of the internal timebase reference. For the highest performance scopes such as the 63 GHz Keysight Z-Series this can be as low as 50 fs but it must be noted that this value is often only valid for fairly short acquisition times. To measure close-in phase noise we need to capture long acquisition times and the intrinsic jitter of the oscilloscope will increase due to its own phase noise.
Noise Floor & Signal Slew Rate
In most cases the first term in the equation dominates the jitter measurement floor. Both signal and oscilloscope vertical noise combine with the finite slew rate of the signal to create apparent horizontal displacement of edges, i.e.: jitter. Thus it is crucial to choose an oscilloscope with as low a vertical noise bandwidth density as possible. A further improvement in jitter measurement floor can be achieved if the oscilloscope also has the ability to limit the bandwidth to an arbitrary frequency. Since the phase noise information is contained within a bandwidth 2*fc we can drastically limit the measurement noise in many cases.
Below (Fig. 2) is a set of phase noise measurements made using a Keysight 8 GHz S-Series oscilloscope. The signal source was a 100 MHz sine wave from an ultra-low phase noise Performance Signal Generator, E8267D. The true phase noise of the E8267D (as verified with an SSA or other suitably low phase noise instrument) is well below the oscilloscope measurements so this enables us to see the measurement floor of the scope.
The oscilloscope bandwidth was adjusted for each measurement as follows:
Blue = 8 GHz, Green = 4 GHz, Red = 1 GHz, Cyan = 200 MHz.
The phase noise floor at 1-10 MHz offsets drops from ~-124 dBc/Hz to -140 dBc/Hz when going from bandwidth of 8 GHz to 200 MHz. This can be explained by the fact that we’re reducing the bandwidth by a factor of 200 MHz / 8 GHz. If the noise of the oscilloscope is fairly flat with bandwidth we should expect a drop of about 10*log10(0.2/8) = -16 dB. This is not the case at all frequencies. At low frequencies the phase noise of the oscilloscope’s internal reference starts to dominate. At higher frequencies we see the limit of the ability to produce a perfect brick-wall bandwidth limit filter at 200 MHz. This means we are still getting some scope noise beyond 200 MHz included in our measurement.
The benefit gained in limiting the scope’s bandwidth is highly dependent on the slew rate of the signal to be measured and the ratio of the signal frequency to the full scope bandwidth.
Noise Floor & Scope Internal PLL/Oscillator
It is often the case with phase noise measurements that low frequency phase modulation is of particular interest. In addition to requiring responsive, deep memory acquisition as discussed in a previous article it is also important to have an oscilloscope with an extremely stable timebase and well-designed PLL circuitry as this will dominate the low frequency measurements.
In the measurement below (Fig. 3) you can see that an older technology oscilloscope (green) has higher phase noise at close-in offsets than the newer technology oscilloscope (blue).
Further improvement of the close-in phase noise might be possible using an external reference clock to the oscilloscope which is cleaner than the internal oscillator. Below (Fig. 4) is a comparison measurement of a Keysight V-Series phase noise floor using the internal oscillator (blue) versus a Wenzel 10 MHz reference (red):
Noise Floor & Sample Rate
Previously I mentioned that the oscilloscope sample rate must be kept high in order to accurately place the edges. It would be nice to be able to reduce the sample rate as it would allow us to use less acquisition points and thus either make faster measurements, increase averaging or go to lower frequency offsets. But we must be careful to make sure the sample rate does not impact our measurement accuracy significantly.
Below (Fig. 5) we can see the impact of reducing the sample rate (bandwidth is maintained at 200 MHz for all measurements) on the phase noise measurement of the same, clean 100 MHz sine wave.
Blue = 1 GSa/s, Green = 5 GSa/s, Red = 10 GSa/s, Cyan = 20 GSa/s.
You can see that eventually reducing the sample rate does impact the phase noise measurement floor. In this case there is not a significant difference between using 20 GSa/s and 10 GSa/s, but below that sample rate there is an increase in the results. The extent of the impact will also depend on the shape & slew rate of the signal edges.
Phase Noise of a Data Signal
Since the oscilloscope uses a clock recovery algorithm to extract the TIE information, an advantage of this approach is the ability to measure the phase noise of data signals. In the example below (Fig. 6) the phase noise of a high speed pattern generator is measured. The only difference in the measurements is the pattern used. Blue is a pseudo-random bit sequence and green is a repeating one-zero clock pattern:
There is some difference in phase noise at high frequency offsets due to the nature of the generator.
To summarize, real-time sampling oscilloscopes – although perhaps not a first choice for phase noise measurements – can be an acceptable choice depending on the measurement requirements. For close-in phase noise measurements (typically less than 1 kHz or so) a dedicated phase noise analyzer or spectrum analyzer will provide a faster, more accurate measurement. However for measuring relatively low cost oscillators and PLL circuits or for wide bandwidth requirements an oscilloscope with a clean timebase and low noise front end may be very capable of making the required measurement. In addition using a real-time oscilloscope has the advantage of allowing you to extract phase noise from a serial data signal if a serial data clock recovery approach is used.
Questions? Visit the Infiniium phase noise forum.