Multiple WiFi cameras on one embedded board: the bandwidth math

You want four cameras watching four positions, one embedded board running the detector, and no new cable between them. That question is arithmetic, not faith: what each camera sends, what the radio carries, and what the detector needs. The sections below walk the numbers for 1080p cameras streaming compressed video at 10-30 fps from a few metres away. You get the bandwidth each piece contributes, worked totals for four cameras, and what a 2.4 GHz and a 5 GHz link actually deliver. You also get the failure modes that pure bandwidth hides, and the wired alternative that machine-mounted cameras should use anyway. By the end you can decide from your own fps and scene numbers whether the wireless build closes or the four cameras need a switch.

Can multiple WiFi cameras feed one embedded board at the same time?

Yes. Four WiFi cameras can stream to one embedded board at the same time, but only when each camera compresses its own video. Four 1080p30 H.264 streams total 16-32 Mbps, which a clean dedicated link carries, while raw camera output at 0.5-1.5 Gbps does not fit at all.

The distinction decides the whole project. A board's WiFi module is one radio on one channel, and everything on that channel shares a single half-duplex pipe. The sum of all four streams is the number that matters. The cameras also need to do the encoding themselves: IP camera modules and streaming camera boards with built-in H.264 or H.265 encoders hand the host a 2-8 Mbps stream instead of a 0.5 Gbps raw one. If any camera must deliver lossless frames for measurement or defect grading, take it off the air from the start. Wireless adds retransmission jitter that no bitrate setting removes.

The working checklist is short: count megabits per stream, multiply by the camera count, add margin, and compare against throughput measured on the channel you will actually use. The sections below carry each number. The systems topic page collects the sibling questions — shared USB buses, frame-rate collapses, host sizing — while this page stays on the wireless arithmetic.

How much bandwidth does one 1080p camera stream use?

Budget 2.5-5 Mbps per 1080p camera at 25-30 fps on H.264 as the starting point. Raise it to 8 Mbps for busy scenes or high quality, and expect roughly half of each figure on H.265, so 1.5-3 Mbps.

Surveillance bitrate guidance starts a 1080p camera at 25-30 fps at 2.5-5.0 Mbps on H.264 and 1.5-3.0 Mbps on H.265, and streaming-quality references use 8 Mbps as the standard 1080p30 H.264 figure. The encoder spends bits on motion and detail, so a static scene sits at the low end and a busy one at the high end. Use variable bitrate with a hard ceiling: the camera idles near 3-4 Mbps and spikes only when the scene moves.

The reason compression is not optional is the raw rate:

Worked through for a mono 8-bit 1920 × 1080 sensor at 30 fps: 1920 × 1080 × 1 × 30 × 8 = 497,664,000 bits per second, about 498 Mbps or 0.5 Gbps. Three bytes per pixel for colour triples that to 1.49 Gbps. Dropping to 10 fps cuts the mono figure to about 166 Mbps. A 4 Mbps H.264 encode of the mono stream is a 124:1 compression ratio (497.7 / 4), and H.265 at 2 Mbps is around 250:1. No radio your board carries moves half a gigabit, so the camera-side encoder is what makes wireless possible at all.

What throughput does a 2.4 GHz WiFi link actually deliver?

A single-antenna 802.11n radio at 2.4 GHz and 20 MHz width reports a 72 Mbps PHY rate. It delivers roughly 30-50 Mbps of real TCP throughput in a clean room at a few metres, and half that or worse with walls or neighbouring networks in the picture.

The PHY number is advertising, not delivery. Real throughput lands near half the PHY rate in good conditions, and the channel is half-duplex: the four cameras and the board's acknowledgements all take turns on one frequency. Embedded boards usually carry a single-antenna module. Plan around the single-stream 72 Mbps figure rather than a two-stream or three-stream headline from the access point's box.

Only channels 1, 6 and 11 stay clear of each other in that band, and everything in a building fights for them: neighbouring networks, Bluetooth, microwave ovens. Bonding two channels lifts the PHY to 150 Mbps but occupies most of the band, so it rarely holds on a populated site. A 5 GHz 802.11ac access point reports 433 Mbps PHY per stream and delivers about 150-250 Mbps at room distance. That is why every recommendation below prefers 5 GHz when the board's module supports it.

Do four 1080p streams fit inside that link?

Four cameras at 4 Mbps each sum to 16 Mbps sustained, and 32 Mbps if all four hit an 8 Mbps peak together. With a 30 percent margin for retransmissions and protocol overhead, the link has to actually deliver about 46 Mbps.

Recording adds no airtime, since each stream still crosses the link once, but it adds storage fast. At a 16 Mbps aggregate the board writes 16 / 8 = 2 MB/s, which is 7,200 MB or 7.2 GB per hour; at 20 Mbps it is 9 GB per hour. The same per-camera arithmetic, scaled to retention days and drive sizes, is what the IP camera storage and bandwidth sizing guide walks through.

The host side needs the same budgeting when cameras arrive over wired buses too. Multi-camera RGB depth bandwidth: size a four-camera host works a four-camera example through host bandwidth and memory.

When does WiFi drop frames even with bandwidth to spare?

WiFi drops frames when retransmissions and contention stack up: a loaded channel adds 10-100 ms of jitter and bursts of packet loss even though the average bitrate fits the link with room to spare.

Contention, not capacity, is the usual killer. Four cameras, the board's acknowledgements and anything else in range all wait their turn on one channel, and a retransmission spike eats the margin in bursts. Cameras spread across 20 ft with a wall or two between them add hidden-node retries on top. The symptom is not a steady slowdown: the detector sees 30 fps on average, delivered as bursts, gaps and occasional frame clusters lost while the encoder resynchronises after packet loss. Timestamps wander by tens of milliseconds, which detection and logging tolerate.

A vision loop that steers a machine in real time does not tolerate them.

The same collapse-by-budget pattern shows up on wired buses. Camera FPS drops suddenly: exposure, USB link, bandwidth traces it end to end on USB.

Which cameras and hosts work better wired?

If the cameras and the board live on one machine or within a cable run, GigE cameras into a vision controller with PoE camera ports beat WiFi on every number: 1,000 Mbps per port, lossless frames, hardware trigger, and camera power over the same cable.

On a machine, the comparison is not close. Each camera gets its own port, so nothing shares the wire: a single 1080p mono stream at 0.5 Gbps rides one port with headroom, which is exactly why the controllers below list 5+ LAN ports. PoE removes the supply at each camera head, and vision controllers with built-in light channels fire strobes from the same box that grabs the frames. For the camera side, these GigE area scan models cover mono detection and colour work on one switch.

On the host side, the MV-VC5525-256G70-NN lists PoE camera ports, 5+ LAN ports, 8 built-in light channels and 32 GB of memory, so cameras and strobes terminate on one box.

The MV-VC3301H-128G66 covers smaller builds with 5+ LAN ports, 3-4 USB 3 ports, 4 built-in light channels, 8 GB of memory and an Intel Core i3 processor. It has no PoE, so camera power comes from injectors or a small PoE switch.

All 2 share: Humidity 20% RH to 95% RH (no condensation); Certification CE, KC. Values as printed in each product's data sheet; open a part for its drawing and full specifications.

The area scan cameras finder filters by interface, resolution and shutter, and the same per-camera budgeting used for WiFi applies to USB hosts. Multiple USB cameras on one host: bandwidth and power budget works that case number by number.

How do you set the link up so four cameras hold 20-30 Mbps?

Give the four cameras a dedicated access point, lock it to the quietest channel or move to 5 GHz, cap each camera's VBR encoder at your per-stream budget, and measure sustained throughput before the last stream goes live.

1. Put the four cameras and the board on a dedicated access point. Sharing the office network adds contention you cannot see or control.
2. Choose 5 GHz when the board's module supports it. On the lower band, survey first and lock the access point to the emptiest of channels 1, 6 and 11.
3. Cap every encoder: variable bitrate, target 4 Mbps, ceiling 8 Mbps per 1080p30 stream. Dropping to 15 fps roughly halves each bitrate when you need more margin.
4. Keep cameras in line of sight where the layout allows. One drywall wall at 20 ft costs margin; concrete and metal racking cost more than the budget survives.
5. Measure before you trust it: run a sustained throughput test between board and access point for several minutes and require about 46 Mbps before calling the fourth stream safe.

If throughput collapses only when the last camera joins, isolate that stream on its own before re-planning. Three depth cameras drop to low FPS: test one stream at a time shows the same isolation method on wired links.

Which link should you pick for your build?

Pick WiFi when pulling cable is impossible and every camera encodes its own H.264 at 1080p or below: the totals close at 16-32 Mbps on a dedicated link. Pick wired GigE with PoE when the cameras sit on the machine, the result steers equipment, or the frames must arrive lossless.

• Four 1080p30 H.264 cameras, one room away, detection or monitoring: build it wireless on a dedicated 5 GHz access point at 4 Mbps per stream; the 16-32 Mbps total holds with margin.
• Same cameras, crowded lower band only: move the streams to H.265 at 1.5-3 Mbps each or drop to 15 fps and re-measure; if the link still will not hold about 25 Mbps, cable at least the busiest camera.
• Cameras on the machine or within cable reach: GigE into a PoE vision controller from day one; wireless buys nothing there and costs jitter.
• Measurement, small-text reading, surface grading: wire it lossless; compression artifacts sit exactly where your features live.

For the wired side, filter the camera finder by interface and resolution, or run the system configurator to match cameras, lighting and controllers to your layout. Reach the team through contact when the channel survey comes back ugly.