Pathology: Parameter Drift or Jump
Synonyms: frequency drift, non-stationarity
Found by: Statistic Tracking, Voltage Statistic Tracking, RjDj Tracking
- Summary Diagnosis and Suggested Solution Path
- Related Pathologies
- Related Topics
- Sample Files
Summary Diagnosis and Suggested Solution Path
HAL has observed that a statistical waveform parameter, such as Mean or Peak-to-Peak (the exact parameter is noted in HAL's description of the problem), has either been drifting over some number of acquisitions, or has had a sudden significant change in value.
Parameter drift can be associated with a changing operating temperature of either the circuit or the test equipment. Changes in period, frequency and amplitude have been observed over time to directly track the operating temperature of some oscilloscopes. For example, a scope placed under the output of an air-conditioning vent. You may wish to ensure that the temperature of your measurement environment is sufficiently stable.
A sudden change in value may be expected due to an intentional change in signal conditions (e.g. a change in input voltage). If this is the case, you can usually safely ignore this alert, and may wish to mute the Agent if you will be making more intentional changes to the signal.
The other major cause of a jump in value is non-stationarity. This can have any number of causes, and it is beyond the scope of this entry to attempt to enumerate them all.
The Parameter Drift/Jump Agents detect when a statistical parameter, such as Mean or Peak-to-Peak, changes significantly. This change can occur either slowly (i.e. drifting) or all at once (i.e. jumping). These Agents will not create any responses until at least 10 acquisitions have been taken, in order to provide the algorithm with a baseline of data from which to look for an anomaly.
The parameter will be seen to either change slowly vs. time, or vs. acquisition for drift. A sudden displacement of the parameter will occur for a jump. Note that this is for parameters that are "directly measured" on the waveform. For parameters that are accumulated, or computed across many events (typically statistical measurements) that might span a long time interval, the presentation can be quite small or displaced in time. Further, it is possible for the dynamics of the way the waveform is changing to interact with the dynamics in the way the waveform is being sampled, measured or displayed to exaggerate the degree of drift or jumping.
When a Tracking Agent detects either drifting or a jump in a parameter being tracked, a new View will be opened in which the value of the parameter will be plotted versus acquisition to provide the user with a visual indication of the behavior. In the case of a drift, the linear best-fit line will also be plotted.
- The most common cause of a parameter drift is a temperature drift in the signal's environment
- Changes in the noise being carried in the measurement power environment can cause some measurements like jitter to increase or decrease. Many labs are "hot" from 10am to 2pm and have a level of noise before and after that time that result in observably quieter measurements.
- This Agent can be triggered if you have intentionally changed signal parameters since the last acquisition (e.g. if you are sweeping Vdd), but did not clear Agent history by clicking on 'Clear History' on the HAL tab. Using 'Clear History' will restart the 10 acquisition learning period before drifts or jumps will be noted.
- The most common cause of an actual sudden change in a parameter is non-stationarity in the signal. Since non-stationarity is by its nature transitory, another Jump will quite likely be detected on the next acquisition, as the signal returns to its normal state.
- Even very clean signal sources can exhibit periods of non-stationarity. They can be brief and infrequent, or consistent.
In devices that regenerate a clock (e.g. PLL's, DLL's, etc), if a drift is sufficiently slow, the device will generally be able to track along with the change, and there may be no noticeable impact.
In parallel regenerative systems (e.g. a PLL type device in at least one branch of a clock-powering tree), the direction of drift along one leg of the tree can be retrograde to a clock emerging from a parallel branch. When those clocks create synchronization with parts of the circuit that communicate (such as the up and downstream state devices of a logic segment), a synchronization failure is possible.
Non-stationarity is a particular problem when attempting to make RjDj measurements, since the math behind RjDj separation does not allow for the existence of non-stationarity. The non-stationarity will often manifest as "BER blooming".
Other anomalies with similar presentations.
- Non-Stationarity - A Jump in a parameter may be associated with non-stationarity, an abrupt (and often short-lived) change in some fundamental characteristic of the signal. Non-stationarity may also create effects such as a change in the number or ordering of peaks in a histogram display. Such a change will be flagged by the Non-Stationarity Agent, and therefore may point to the same underlying issue as a Jump condition.
Maintaining your measurement environment to within a small temperature swing is important to making accurate and repeatable measurements. You should avoid placing the scope/DUT in the direct path of heating/cooling vents or in direct sunlight, as such areas will usually exhibit a much larger temperature swing than other areas.
When making measurements across very long timescales, it is also possible for the measurement power environment to change through the course of a day. ASA has been called out on some consults to help reduce sudden increases in jitter that ultimately were caused by: a vacuum cleaner in the room next to the lab in Texas, a freight elevator motor on the other side of the wall from the power distribution point for a lab in Oregon, cell phone induced failures in Oregon, and test equipment adjacent to the DUT and it's own test gear being turned on or off or changed on numerous occasions.
Non-stationarity has a large number of possible causes, both within the measurement environment and within the DUT itself. We will not attempt to list all of the possibilities here.
There are no files available demonstrating Parameter Drift Or Jump. If you would like to contribute some sample data, please click "contribute" below.
If you don't have M1, you can download a trial copy here .
Would you like to contribute?