Undershoot — Control Systems/Pulse Metrics

Undershoot

Control Systems / Pulse Metrics

Reports how far the waveform fell short of, or dipped below, the state level it was heading for, as a percentage of the amplitude – MATLAB's undershoot:

us = (landing state level − min of the aberration region) ÷ (high − low) × 100

It reads the same IEEE Std 181-2003 Sec 5.3.5 aberration region as Overshoot – from the sample the transition settles on, for Seek Factor transition durations or until the next transition – and reports the other end of it.

Ports

• u – the bilevel waveform to measure. Scalar – see Notes.
• us – percent undershoot of the most recent closed aberration region, held until the next one closes. Scalar. A waveform that settles without dipping reports 0 or less.

Parameters

• Low State Level – the value the waveform sits at in its low state. 0 by default.
• High State Level – the value it sits at in its high state. 1 by default. Every threshold below is a percentage of the distance between these two, so they must differ.
• State Level Tolerance (%) – how far into the amplitude a sample must be before it counts as having settled into a state. 2 by default, the value IEEE Std 181-2003 uses. Samples between the two boundaries belong to no state and are treated as part of a transition.
• Lower Percent Reference Level – the level a transition is timed FROM on a rising edge and TO on a falling one, as a percentage of the amplitude. 10 by default, as IEEE Std 181-2003 specifies.
• Upper Percent Reference Level – the other end of the same measurement: timed TO on a rising edge and FROM on a falling one. 90 by default. ⚠ These two also decide where a transition is DATED, because the mid reference level is their mean – the same rule MATLAB's transdurs applies, so an asymmetric pair such as 20/70 moves the transition instant to 45% rather than leaving it at 50%.
• Seek Factor – how long the aberration region is, measured in transition durations. 3 by default, as IEEE Std 181-2003 specifies and as MATLAB's SeekFactor does. A longer region catches a slower ring at the cost of running into the next transition, which truncates it anyway.
• Sampling Time (s) – zero or less inherits the solver's rate; a positive value runs the block at that period.

Code export

All ten targets: Python, MATLAB, Java, Rust, C, C++, VHDL, Verilog, SystemVerilog and PLC Structured Text.

Every threshold is structural and is inlined into the comparisons at export time rather than exposed as a tunable parameter – they follow from the state levels and the percentages, and the emitted core carries the numbers rather than the percentages. Re-export after changing any of them.

⚠ The three HDL targets are simulation-only. They carry this block's arithmetic in real rather than Q16.16, quantising only at the port boundary, for a reason specific to what it measures: an instant is a sample index plus a fraction, and a 16-bit integer part runs out after 32767 samples – five and a half minutes at a 10 ms rate. The comparisons and the interpolation are exact in real, which is what makes the exported core agree with the simulation; a synthesizable variant would have to carry the index some other way and is not what this block offers.

Simulink bridge

No equivalent (Support::None). Measured rather than assumed: a find_system sweep over 22 DSP System Toolbox and Simulink libraries on R2026a finds no block matching duty, pulse, rise, fall, slew, overshoot, settling or state level – the whole Pulse Metrics group is MATLAB functions with no Simulink counterpart. The bridge reports this block rather than dropping it silently, and it therefore has no parity testbench. Code export verification still covers it across all ten languages.

Notes

• Stateful, and discrete by nature (setDiscreteOnlyBlock(true)): the tracker advances once per sample.
• A measurement appears one sample after the transition that produced it, in the live run and in all ten exported languages alike. A hardware block cannot read back a value it computes in the same clocked interval, so publishing immediately would put the three HDL targets one sample behind the other seven on every edge; the delay is uniform instead, and deliberate.
• Crossing instants are interpolated between the two samples that straddle a reference level, exactly as IEEE Std 181-2003 Sec 5.3.3.2 specifies and MATLAB implements – so a measurement is not quantised to the sampling period.
• The output is 0 until the first measurement exists, and holds its last value between them. A run that never settles into both states produces no transitions and the output stays at 0.
• The state levels are PARAMETERS here, not estimated from the data. MATLAB's pulse-metric functions call statelevels to guess the two levels from a histogram of the WHOLE signal when you do not supply them; that is a batch operation over samples a streaming block has not seen yet. This block takes the levels the way those same functions do when you pass 'StateLevels' explicitly.
• Scalar only. One waveform and its own tracker; wire one block per channel.
• No state space. The block is not linear in its input – it is a sequence of comparisons – so it carries none and model reduction correctly declines to merge it.
• The value appears when the region CLOSES, not when the transition completes – the region is defined forward in time, so its extreme is not known until it ends. The region ends at the seek deadline, or at the next transition if that comes first.
• ⚠ A truncated region ends one sample earlier than it looks. When the next transition cuts the region short, MATLAB's getPostshoots ends it at the sample before the last settled one – its index guard decrements the truncation the same way it decrements the deadline. This block does the same. Measured against R2026a on a 1.3 Hz sine: taking the last settled sample instead is 0.91 percentage points out, and taking the sample before it agrees to 3e−15.
• Both aberrations are computed from one region. The extreme above and the extreme below the landing state level are tracked together, so this block and its sibling read the same window rather than two windows that could disagree.