Browse Prior Art Database

Method for Quantification of Waveform Differences Disclosure Number: IPCOM000125449D
Original Publication Date: 2005-May-31
Included in the Prior Art Database: 2005-May-31
Document File: 4 page(s) / 54K

Publishing Venue



Hand comparison of two waveforms is a tedious task that can take up much time. Waveforms that vary significantly on one axis (e.g., voltage) over a small range on another axis (e.g., time) are especially difficult to compare, due to the number of data points usually found therein and the fact that small differences are difficult to measure and quantify by hand. In the case where many waveforms are involved, numbering in the hundreds or thousands, hand comparison becomes almost impossible due to the time needed to properly analyze differences. For this reason one would turn to automation in order to compare the waveforms in a reasonable amount of time. It is not enough to compare two waveforms, point by point, even in an automated fashion; small offsets on one axis between the two waveforms would yield a difference in every data point, for example. Accuracy of the comparison in our case was seen to be the overriding requirement for the automation of the comparisons performed. The problem to be solved was to find a way to compare each waveform very accurately while taking into account (and correcting for in the analysis) macroscopic differences between the "golden" and "test" waveforms. Additionally, a method was required to enable the reporting of these differences in a meaningful fashion. A system of comparison was devised which allowed the automated comparison and quantification of differences between two waveforms which gives specific report as to how well the two waves agree with each other, and a quantification of how much the two waveforms differ and where the largest difference lies within the test waveform. A secondary problem was encountered when a "golden" waveform was not available, but we had a number of related waveforms in a group. In this case, an analysis of trends in the waveform group was required, and automation again was seen as the most efficient way to accomplish the task at hand. Our analysis would need to spot differences or anomalies in the trend within the waveform group and report these in a meaningful fashion. Accuracy was again an overriding requirement in this case. A system of comparison was devised which allowed the automated comparison and quantification of differences within a group of related waveforms, giving specific report as to which wave(s) in the group did not follow the overall trend and identifying the point(s) at which they stood out from the group.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 35% of the total text.

Page 1 of 4

Method for Quantification of Waveform Differences

Our design methodology generates a large number (1000s) and variety of waveforms that need to be validated. The current method was to visually inspect these waveforms which is very tedious and labor intensive. We desired a reliable means of automation to accomplish this validation. We do not know of any alternative solutions.

     This invention allows for complete automation of the validation process saving many weeks of manual visual inspection and validation. Our waveform validations are of two types. The first is to characterize the differences between two waveforms. In this case one waveform is considered correct. So we have established a 6-point characterization which gives us a standard for how close the waveform under test is to the "correct" waveform. This method is extremely useful in development as we are able to run automated comparisons of databases of simulation waveforms. Our waveforms are from circuit simulation and are typically voltage versus time. The 6 point characterization was designed to quantify both y axis and x axis differences in detail. Prior to our invention, a visual inspection of 1000's of waveforms was necessary in the development of ~600 models.

     The second type of comparison is for the case when we do not have a correct waveform. We have established a 3-point characterization to determine the validity of a group of waveforms in an automated manner. This characterization has proven extremely useful as we develop hundreds of models that we run characterizations such that each model generates a group of waveforms (~20-50 waveforms for each model). This 3-point characterization validates the group of waveforms for each model in an automated manner. Prior to our invention this was achieved through visual inspection of over 2000 waveforms.

     Method for waveform comparison to a known standard is based on calculating six metrics. Three of the metrics (M1-M3) quantify changes in the values in the y axis. The other three metrics (M4-M6) focus on measuring the differences in the area under the curve and are very good at detecting even small amplitude changes. Based on our applications, an exact match would have the following values: M1=100%, M2=0%, M3=0.0, M4=0.0%, M5=0.0% and M6=0.0%. The six metrics are as follows: M1 represents the average of the absolute value of the relative errors M2 represents the maximum error relative to NF (normalization factor)

M3 represents the maximum absolute error M4 represents the Relative Local area error M5 represents the Relative Global area error M6 represents the Maximum Absolute Local error

Variable Definition for M1, M2, M3:

Delta_Y(i) = Y1(i) - Y2(i)

N: Total number of point.

     NF: Normalization Factor for the amplitude variation (Y) (for example: logic swing)

M1: 100 * { 1 - [ Sum of (Delta_Y(i))/ (N * NF) ] }

M2 : 100 * { Max of (Delta_Y(i)) / NF }

M3: Max of (Delta_Y(i))


Page 2 of 4

Variable Definition for M4, M5, M6: