In this case, let us assume that the switching distribution is uniformly distributed in time. The resulting waveform for this ideal case would take on the form of white noise. In the case of the NCL PRNG, the switching distribution is not completely uniform. As seen the distribution takes on a periodic behavior corresponding to the generation of the pseudo-random numbers.Thus, the resulting substrate noise waveform reflects the shapeof the switching distribution. Note that in this experiment, thereare a total of running the same algorithm. Towards test-driven software defined networking Since the rate at which each individual block operates is slightly different, the superimposed ringing in the substrate may exhibit some additional amplitude modulation.Analysis of Substrate Noise in Frequency Domain As revealed in the time domain analysis, the reason the skirted peaks in represent the equivalent operating frequency of the PRNG is because the switching distribution happens to reflect the generation of the pseudo-random numbers. Although the connection between the equivalent operating speed and frequency peaks in the spectrum should not be true in general, it will be seen later that most structured architectures will exhibit some repetitive properties that will produce similar frequencyfeatures. Another characteristic of interest in is the apparent shape of spectrum. This shape is much more noticeablein the secondary tones. In both spectra, the secondary tones form an envelope that slightly peaks at about. Using Fourier analysis, Towards test-driven software defined networking it can be inferred that the difference between these secondary tones reflects the general periodicity of the time domain waveform while the envelope indicates the oscillating component between each noise spike. In this case, the general periodicity should be the result of the sequence length of the PRNG, which repeats approximatel Towards test-driven software defined networking y every the noise spectrum from the NCL PRNG shows the expected behavior with skirted tones at about and its harmonics.However, the interval between the secondary tones for the CBL PRNG is actually twice the sequence length frequency. Further analysis reveals that although the sequence of the data rep Towards test-driven software defined networking eats every cycles, there is a strong repetitive component inthe transition distribution that happens to repeat every cycles. This frequency component Towards test-driven software defined networking is not as prominent in the NCL PRNG since its asynchronous nature also skews the generation of the pseudo-random numbers in the time domain. So although the pseudo-random stream itself may have a strong repetitive component, the time required for each code-to-code transition is unique.Chip-on-Board Package Measurements the substrate noise measurements of the CBL and NCL PRNGs when the die was mounted directly on the PCB as a chip-on-board COB. Here, the package parasitics are simply small inductances from the bond wires. As seen in the figure, the substrate noise generated by the CBL PRNG is again much greater than that caused by the NCL PRNG. In terms of numerical values, the RMS noise voltage from the NCL PRNG was lower than that from the CBL PRNG. This difference is very similar to the improvement seen with the package. The actual enhancement offered Towards test-driven software defined networking over the high inductance packaging was for the NCL PRNG. Similar to the high inductance case, there is a train of repetitive noise spikes corresponding to the edges of the CBL clock. In this case, however, there is a much more pronounced variation in the peak magnitudes of the noise spikes. This observation suggests that switching noise is more dominant in low inductance The corresponding frequency spectra of the COB version of the CBL and NCL PRNG substrate noise measurements is Again, these spectra are very similar to the measured spectrum for the high inductance package.