Digitize analog camera video for testing and measurement

You need to test a camera from a medical imaging system: a 1K × 1K CCD whose only accessible output is analog video, and you want that image back in digital form, ideally Camera Link, with as little disturbance to the signal as possible. Every frame grabber you have found is digital-only, and that is correct — none takes analog input. This guide gives the sampling-rate arithmetic for the 1K × 1K stream, the two capture paths that work on a test bench, the realistic way to produce Camera Link output, the signal-quality budget for one conversion, and the checks that keep the converter from skewing your measurements.

Can a machine-vision frame grabber accept an analog camera input?

No modern machine-vision frame grabber takes analog video: every board in the frame grabbers category samples a digital interface — GigE, 2.5GigE, 10GigE, CoaXPress, Camera Link, XoFLink or USB3 — so the analog-to-digital conversion has to happen in a separate stage ahead of any grabber.

The reason is electrical, not commercial. These boards terminate Ethernet pairs, coax or Camera Link's differential pairs and decode pixels that are already digital; none of them carries the video ADC and sync separator an analog front end needs. There is no catalog adapter for the gap either, because Camera Link's low-voltage differential signaling only exists after conversion — attempts to cable an analog camera straight into a digital grabber fail before an image forms, as why analog cameras fail on Camera Link frame grabbers explains.

So budget a digitizer stage: a device that samples the waveform and recovers sync, with a frame grabber behind it only if the requirement demands one. Every board in the category sits under the accessories finder.

What does the analog output of a 1K × 1K CCD camera contain?

Treat the output as one video waveform — 0.7 V of active video with sync into 75 Ω — whose line rate and pixel clock you must measure, because a 1K × 1K progressive camera is non-standard and ignores the 640 × 480 RS-170 timing most capture gear assumes.

Standard video sets the contrast. An RS-170 camera runs 525 interlaced lines at about 30 frames/s: a 63.5 µs line, roughly 52 µs of active video, and a pixel clock near 12 MHz for square pixels. 625-line CCIR gear is the 50 Hz equivalent. Composite color buries chroma inside the luma band — the origin of cross-color artifacts — while a monochrome CCD output has no chroma to corrupt.

Your camera runs about three times those clocks: 1024 progressive lines at 30 frames/s needs roughly 39 MHz including blanking. That is why broadcast-pattern capture hardware misbehaves on it. Put a scope on the output before buying anything, and measure five things: line period, active-video time, sync type (embedded in video, composite sync on a separate conductor, or discrete H and V), polarity and amplitude. Sync type decides which digitizer will lock; pixel clock decides its sample rate, which is the next number to compute.

What sampling rate keeps the conversion from losing image information?

Sample at twice the camera's pixel clock, and no lower: for a 1K × 1K progressive camera at 30 frames/s the active pixel rate is about 31.5 Mpx/s, the clock including blanking lands near 39 MHz, and a converter running 80 MSPS or faster digitizes every pixel without aliasing.

The 0.8 is an illustrative duty cycle; blanking typically consumes 15-25 % of a line, so divide your scoped active time by the line period for the real figure. The Nyquist logic underneath is not negotiable: sampling below twice the highest video frequency folds genuine image detail into false patterns that no software removes afterwards. The front end also has to pass the video bandwidth, about half the pixel clock — call it 20 MHz here.

Frame rate is your cheaper lever. At 15 frames/s the same camera needs a 19.7 MHz clock and about 40 MSPS, and a slower bench run still measures every property except speed. Choose the rate deliberately instead of accepting what a capture card happens to do.

Which hardware path should you use to digitize the video?

Use a purpose-built video digitizer in front of the PC — a capture board or decoder chip with sync separation — and keep the analog path to one cable and one conversion; build a PC-based analog-to-Camera Link converter only when downstream equipment must receive Camera Link.

Path A is the test-bench default: capture software takes the stream into memory as plain frames, and wiring it into your measurement code is the same problem as any camera SDK — integrating an analog frame grabber with vision software walks that stage. The trap is consumer composite capture: those devices expect 525/625-line interlaced standards at a fixed clock and will refuse to lock, or tear, a 1K progressive stream. Buy a capture device documented for discrete-sync progressive operation with a pixel-clock range covering yours; triple-input HD video decoders with automatic sync detection, the class inside broadcast-grade capture cards, are what to search for.

Path B is that same front end plus an output stage, covered next.

How do you get the digitized stream back out as Camera Link?

Generate Camera Link from the captured data in software through a board with data-forwarding or output capability, or buy a digitizer that already outputs Camera Link; then any Camera Link frame grabber, such as the MV-GC1102IOL, reads the stream like a native camera.

First, the format you are generating. Camera Link packs a 28-bit word per pixel tap: up to 24 bits of pixel data, three sync bits marking frame, data and line validity, and one spare. A Base channel clocks pixels at up to 66 MHz and carries only active pixels — the sync bits gate blanking off, so it is not transmitted. Your stream needs 31.5 MHz against that 66 MHz ceiling, with 8-bit pixels occupying 8 of the 24 data bits: one Base channel takes it with 2:1 headroom, per the Camera Link format summary.

The output side is the scarce part. Frame grabbers are receive-only by design, so generating a stream takes a board documented for Camera Link output or data forwarding, driven by software that writes each captured frame into it. That is the honest shape of an analog-to-Camera Link converter: one capture front end, a little software, one output board. Question the requirement while you are at it — if Camera Link matters only because a downstream instrument speaks it, the bridge earns its keep; if the goal is simply digital data, Path A ends the project one stage earlier.

The receiving end of that stream is an ordinary Camera Link frame grabber:

If the wired chain comes up silent, the usual culprits are tap configuration, sync polarity and cabling; the signal checks for a frame grabber that cannot detect its video source run them in order.

How much signal quality does one clean conversion cost?

A single 8-bit conversion costs almost nothing you can measure: an ideal 8-bit quantizer limits SNR to 6.02 × 8 + 1.76 ≈ 50 dB, and its code width on a 0.7 V video swing is 2.7 mV, about what the camera's own 8-bit chain resolves.

The camera sets the reference. Its electronics were designed to hand a grabber 8 bits of 1 V p-p video, so a converter that resolves the same 256 levels across the active 0.7 V preserves everything the camera resolves. A 10-bit front end adds no gray levels the sensor never produced; it adds headroom for gain and offset without consuming codes.

Damage lives downstream of the ADC. Undersampling folds high-frequency detail into false low-frequency patterns. A high-impedance input reading a 75 Ω source lands at roughly double the intended level, and unmatched long cables add reflections. Lossy compression moves pixel values after the fact. One ADC, one cable, correct termination and uncompressed storage make the conversion transparent at the camera's own depth.

How do you verify the bench measures the camera and not the converter?

Calibrate the chain end to end before trusting any measurement: feed a known test waveform or a uniform target, confirm the digitized gray-level span matches the camera's specification, and only then attribute what remains to the camera.

Work down this list once, then spot-check after any hardware change:

1. Scope the waveform. Line period, active time, sync type and polarity, amplitude into 75 Ω — every downstream setting keys off these.
2. Baseline the converter. Feed a linear ramp and log the code error; that error belongs to the digitizer and comes off every later measurement.
3. Flat-field it. Uniform target, histogram mean and spread against the camera's specification; extra spread is the chain's contribution.
4. Run a gray wedge. Codes must step monotonically at about 2.73 mV per level; a stuck or non-monotonic code marks a front-end fault.
5. Check geometry. Exactly 1024 × 1024 active pixels with straight edges. Torn or displaced lines implicate sync capture, not the sensor — the same symptom as an analog camera image split in two by a cable fault, and scanline distortion traced to capture settings covers the timing variants.
6. Store raw. Uncompressed frames only; saving camera frames to disk for offline testing shows the capture side.

Once wedge and flat-field agree with the camera's specification, what remains is the camera — the measurement you were hired to make.

When should you replace the camera instead of digitizing its output?

Replace the camera when the test requirement is permanent or production-facing: a modern digital camera removes the converter, and a 1024 × 1024 Camera Link, GigE or USB3 camera with the same lens mount drops into a digital capture chain without adaptation.

Keep the converter for the cases that genuinely need it: the camera is a spare for installed medical systems and cannot be substituted, the qualification program pins the exact model, or this is a one-time characterization of hardware you do not keep. One clean ADC is then the only route to the analog signal, and everything above applies.

For recurring testing, ask what consumes the stream. If the answer is your measurement software, Path A was always the answer. If the camera itself is aging, re-cameraing costs less than maintaining a custom converter indefinitely: choosing between analog and digital cameras frames that decision, and replacing an obsolete analog NTSC frame grabber with USB3 Vision records a complete swap. The replacement camera's interface then sets the receive card:

The MV-GE1104 is a 4-port GigE frame grabber for GigE Vision cameras, the MV-GT1102P a dual 10GigE frame grabber for higher-rate models, and the MV-GU2204 a quad USB3.0 adaptor for USB3 Vision cameras.

At 8 bits the digitized stream runs about 31.5 MB/s — 31.5 Mpx/s times one byte — which any current host absorbs, including a fanless industrial mini PC when the bench has no room for a tower. Size the receive side at the frame-grabber step of the configurator; the surrounding capture-chain articles sit on the systems topic page, and if you send the camera's timing sheet through the contact page, we will name the board.