Calculating Global Overlap Windows in Sequential Row Readout Image Sensors

Raster

Sequential row architecture scans a photodiode array line by line, introducing a fixed time lag between when integration starts on the first row and when it starts on the last active row. Every row integrates charge for the exact same duration, but the physical charge collection takes place in different time slices. An electronic rolling shutter runs this sequence by sending a reset pulse to row zero, waiting out the programmed integration delay, and transferring photo-generated electrons to the floating diffusion node for analog-to-digital conversion before moving down the die.

Working out this spatiotemporal skew requires looking at the horizontal line period, usually labeled 1H or HMAX in register maps. That period is the sum of active row digitization time and the horizontal blanking interval. Horizontal blanking leaves room for analog settling times, correlated double sampling conversion cycles, and serializer line formatting.

If an optical sensor has 2160 active vertical lines and runs with a line time of 14.8 microseconds, reading out the entire array takes 31.968 milliseconds.

Under a 74.25 megahertz master pixel clock, a 1080p sensor configuration registers exactly 19.2 microseconds of horizontal line progression across 1125 total row clocks.

Because the reset and readout scans traverse the pixel array at the exact same velocity, the integration windows for all rows tilt diagonally on a time-versus-row position plot. Reset starts the exposure; readout ends it. Vertical blanking between successive frames delays the next frame’s reset pointer, preventing continuous back-to-back integration unless specific overlapping exposure modes are active in the timing generator.

Line Timing Displacements

Total frame readout duration sets the baseline time displacement across the sensor surface. When fast-moving objects cross the field of view under continuous light, straight vertical lines shear into slants. A vertical post moving horizontally across the optical axis appears further ahead in earlier rows and further behind in later ones, producing spatial distortion proportional to target velocity divided by total frame readout speed.

Calculations of geometric fidelity separate the sensor into discrete timing regions. Three structural parameters dictate the row-to-row progression across modern CMOS dies:

• Horizontal line period sets the exact time required for a single row reset, correlated double sampling, and ADC ramp conversions.
• Array row count determines the multiplier applied to the line period to yield total array readout latency.
• Vertical blanking overhead sets the idle time between reading out the final row of one frame and resetting the first row of the next.

Optical systems running under continuous illumination cannot correct this skew electronically without post-processing interpolation. Eliminating motion tilt at the sensor level requires locking external illumination to a state where every row on the array accumulates charge at the exact same time.

Overlap

Coincident exposure across all pixel rows happens when the exposure duration is longer than the array readout time. Since row zero starts integrating first and row N starts last, full array concurrency cannot begin until row N fires its reset. That overlap ends the moment row zero enters readout.

The window bounded by the reset of the last row and the readout of the first row forms the global overlap window.

Let T_row represent the horizontal line time, N_rows the active vertical line count, and T_exp the programmed exposure duration. The total readout duration of the active array equals N_rows multiplied by T_row. A positive coincident exposure duration, designated T_overlap, exists under one mathematical condition:

T_overlap = T_exp – (N_rows T_row)

If T_exp is equal to or less than total array readout time, T_overlap turns zero or negative. When negative, row zero finishes readout and closes its integration window before the last row even begins exposure reset. Firing a light pulse during a negative window exposes only a strip of the sensor, leaving the remaining top or bottom rows completely dark.

Which Exposure Parameters Dictate the Coincident Interval?

Integration registers set the time between line reset and line transfer in integer multiples of the horizontal line period, accompanied on high-resolution sensors by fractional clock offsets. To get a 2.0 millisecond overlap window on a 3000-row array with an 8.0 microsecond line time, array readout takes 24.0 milliseconds, so the timing controller requires a programmed exposure of 26.0 milliseconds.

Lengthening the exposure opens the time window needed for the strobe pulse, but it also exposes the die to continuous background light. The moving target sees an effective exposure equal only to the duration of the light pulse ~ assuming ambient light is negligible. Meanwhile, the photodiodes continue gathering dark current and stray photons across the full 26.0 milliseconds, lowering the signal-to-noise ratio compared to a true global shutter pixel with dedicated memory nodes.

A standard optical bench protocol establishes that strobe illumination intensity must surpass steady-state ambient illumination by at least forty decibels to suppress ghost edges.

Miscalculating the overlap boundary clips the strobe pulse against either the initial reset sweep or the trailing readout sweep. If the light pulse bleeds into row zero’s readout, a bright-to-dark gradient bleeds across the top of the image. If the pulse fires before the reset sweep reaches row N, the bottom lines receive no light at all, leaving solid black bands across the bottom of the frame.

Strobe

Solid-state light sources can deliver high-flux, microsecond pulses that freeze motion during the overlap window. LEDs and pulsed laser diodes serve as the primary sources for synchronized machine vision light. A trigger controller timing the driver ensures the flash fires strictly within T_overlap, isolated from the rolling transitions at the top and bottom of the frame.

Sensors provide dedicated output pins ~ often labeled STROBE, FLASH, or TIMER OUT ~ driven by internal logic comparators. These signals go high once the internal counter passes the last row reset and drop low just before row zero readout begins. Optocouplers or digital isolators stepping this CMOS logic up to industrial 24-volt lines introduce propagation delays that eat into the available overlap window.

Execution Workflow for Coincident Pulsing

Synchronizing the optical pulse with sensor integration registers demands a deterministic execution sequence:

1. Read sensor register maps to pull horizontal line time and active row counts.
2. Calculate baseline readout time by multiplying active rows by horizontal line time.
3. Determine required flash duration based on target velocity and allowed pixel blur.
4. Add safety margins for optoisolator delays and emitter rise times to set total overlap.
5. Combine total readout time and total overlap to get the minimum exposure register value.
6. Write this value into the coarse integration registers before arming the trigger.
7. Set up hardware output pin polarity and lead-in delay registers.

Problems arise when thermal throttling in high-power LED drivers stretches pulse rise times beyond tolerance. If a driver exhibits an inductive current ramp of 40 microseconds, the optical emission curve pushes toward the back end of the window. Temperature swings also shift the forward voltage of LED arrays, driving power supplies into compliance limits and dragging pulse tails directly into the readout phase of early rows.

When optical pulses collide with rolling readout fronts, spatial illumination gradients corrupt pixel values across the top thirty percent of the active matrix.

Rolling-shutter strobe overlap can reach performance comparable to specialized five-transistor global shutter pixels. That comparison, however, ignores the accumulation of ambient light over the extended exposure interval.

Margin

Reliable synchronization between sensor timing logic and external light drivers requires subtracting hardware timing uncertainties from the theoretical overlap window. Standard equations assume zero jitter, instant transistor switching, zero propagation delay, and perfectly square optical pulses. Actual machine vision hardware introduces jitter and phase quantization that eat away at that window.

Clock domain crossings between the pixel clock and an external microcontroller or FPGA add a quantization uncertainty equal to one horizontal line period. Sensor strobe output signals usually update only on line-start boundaries rather than arbitrary master clock cycles, shifting the physical strobe pulse by up to plus or minus one T_row relative to the photodiode integration fronts.

Is Jitter Suppression Feasible under Variable Frame Rates?

Asynchronous triggers from optical encoders or proximity sensors vary the frame blanking interval dynamically. If a trigger arrives while the sensor is serializing data, the timing generator holds the next reset sweep until internal logic finishes the active line or frame transfer. This variable delay, known as trigger phase jitter, disrupts alignment between external flash controllers and internal sensor counters.

Accommodating these timing variations yields an effective usable overlap window, designated T_usable, calculated through explicit subtraction of edge margins:

T_usable = T_overlap – (T_margin_lead + T_margin_trail)

The leading margin covers sensor trigger jitter, line quantization, and driver turn-on delay. The trailing margin accounts for optical fall times, phosphor persistence, and digital turn-off delays. High-output white phosphor LEDs have long yellow emission tails that linger hundreds of nanoseconds after current stops, whereas direct bandgap red, green, or infrared LEDs drop to zero photon emission within tens of nanoseconds.

System designers balance these timing budgets against target motion. For an inspection system where one pixel of motion blur is acceptable on an object moving at five meters per second with 50 micrometers per pixel resolution, the maximum optical pulse length is 10 microseconds. If timing uncertainty totals 25 microseconds, the overlap window must span at least 35 microseconds ~ which forces the sensor exposure far beyond simple array readout time.

A reliable installation balances illumination energy against timing jitter without pushing the pulse edges toward the scan boundaries.

Extinction

Freezing motion with pulsed light on sequential sensors fails when ambient factory light enters the optical path. Unlike charge-domain global shutter pixels that store signal under light-shielded transfer gates, standard four-transistor rolling shutter photodiodes collect light continuously through the full exposure duration T_exp. The target is illuminated by the flash for only a few microseconds, but the pixels gather background light over the entire frame readout plus the overlap window.

This limits the effective shutter extinction ratio. Defined in standards like EMVA 1288, extinction ratio measures a sensor’s ability to reject light outside the active exposure window. When using long integration times to force an overlap window, the true optical extinction ratio equals the integrated flash photon flux divided by the total ambient flux accumulated over the complete line integration period.

Rejection Metrics in Continuous Ambient Fields

Engineers calculate contrast loss by comparing illumination energy delivered during the flash against background energy accumulated over the rest of the exposure interval. Bright factory light washes out low-contrast defects and overlays a blurry ghost image on top of the flash-frozen primary image. Four factors dictate how severe this leakage gets:

• Optical filter bandwidth limits background photons to a narrow spectral band centered on the emitter wavelength.
• Continuous lux level sets the baseline ambient flux hitting the photodiode outside the flash pulse.
• Array readout latency determines how long photodiodes collect unwanted background light.
• Peak pulse irradiance lifts the target signal above the accumulated ambient photon floor.

Installing narrow bandpass filters matched to the LED spectrum cuts ambient light significantly. Pairing a 630-nanometer LED with a ten-nanometer full-width-at-half-maximum bandpass filter blocks up to ninety-five percent of ambient fluorescent or daylight illumination, preserving image contrast even with thirty-millisecond integration windows.

Dark current adds an internal noise source on top of ambient light. Silicon photodiodes generate thermal electrons continuously, doubling roughly every six to eight degrees Celsius. In industrial environments where camera enclosures reach fifty-five degrees Celsius, expanding exposure time from two milliseconds to fifty milliseconds to create overlap raises dark current shot noise by a factor of five, eroding dynamic range in subtle inspection areas.

Under standard factory floor conditions, an unshielded rolling sensor operating with a thirty-millisecond integration window loses up to twelve decibels of signal-to-noise ratio to ambient contamination.

Choosing between low-cost rolling shutter sensors with strobe overlap and true global shutter dies comes down to balancing silicon costs against lighting hardware. While a rolling shutter chip costs less on a bill of materials, the required high-power drivers, heatsinks, bandpass filters, and light shielding often erase those savings in system overhead. Whether improving back-illuminated global shutter yields will close the die cost gap faster than illumination hardware prices fall remains an open question.