Robot Interconnect Budget

Sensors

One row per sensor type. Reduction divides the raw rate (1 = raw).

Links

One row per link type. Copies spread the load evenly. Usable = rate × lanes × payload ratio × copies.

Load per link

Compute uplink and thresholds

BER, eye and test time

Dual-Dirac jitter budget and the bits needed to prove a BER at a confidence level.

• Baud = line rate / bits per symbol (NRZ 1, PAM3 at most log₂3 = 1.585, PAM4 2, Rohde & Schwarz GFM355). UI = 1 / baud. Nyquist = baud / 2.
• BER = 0.5 erfc(Q / √2), solved for Q numerically. Check: Q(1e-12) = 7.034, matching Keysight Table 2 (7.0) and Renesas AN‑815 (2Q = 14.069).
• Tj = DJ + 2Q × RJ rms. Eye width = UI − Tj. The 802.3ae draft uses 13.73 × RJ at 1e-12; the tool uses the Gaussian 2Q.
• Zero errors: N = −ln(1 − CL) / BER bits (TI SLAA582). With k errors allowed: the N where the Poisson probability of k or fewer errors at mean N × BER equals 1 − CL (Maxim, Lightwave). Time = N / bit rate. A link taken from the list that runs faster on the wire than its MAC rate (FEC Ethernet) uses the MAC rate, since its target BER is defined at the MAC.
• Eye pill: closed at or below 0, tight under 25% of UI (tool threshold).

Interface reference

Every link the tool offers, with the quote each figure came from.

Sensor models

Published figures only. Bits per pixel, bytes per point and ADC width are assumptions you set.

Robot sensor stacks

BER and jitter references

Formulas and assumptions

FAQ

How is usable throughput worked out?

Rate per lane times lanes times a payload ratio. PCIe 1.0, 2.0 and USB 3.2 Gen 1 use 8b/10b, so 0.8. PCIe 3.0 to 5.0 use 128b/130b; USB 3.2 Gen 2 and USB4 v1 use 128b/132b. PCIe 6.0 Flit mode keeps 236 of every 256 bytes. FPD-Link IV carries 6 Gbps of video in 7.55 Gbps. Ethernet rows apply framing at a 1500-byte MTU, and classical CAN uses a derived frame efficiency of 0.47. Where no payload ratio is published (A-PHY, GMSL, D-PHY, CAN FD) the raw rate is used and tagged unsourced.

How much do Ethernet frames cost?

Each frame adds 38 bytes: preamble 7, SFD 1, header 14, FCS 4 and a 12-byte gap. A 1500-byte payload leaves 1500 / 1538 = 97.5 percent for data, before IP and UDP headers. PHY coding such as the 1000BASE-T1 RS-FEC sits below the MAC rate and does not cut it further.

Which robot sensor stacks are included?

Waymo's 6th-generation Driver (13 cameras, 4 lidars, 6 radars), Boston Dynamics Spot (5 stereo camera pairs) and Unitree G1 (one depth camera, one 3D lidar), each from the maker's own page. The makers do not say which link carries each sensor, so link choices and stand-in models are labelled assumptions.

How long must a link run to prove a BER?

With zero errors allowed, test N = −ln(1 − CL) / BER bits. At 95 percent confidence that is about 3 / BER bits, so proving 1e‑12 at 10 Gb/s takes about 5 minutes.

What Q does a BER of 1e-12 need?

Q is about 7.03, from BER = 0.5 erfc(Q / sqrt 2). Total jitter is then DJ + 2Q x RJ rms, so random jitter counts about 14.07 times its rms value.

Why is the camera bitrate an estimate?

Width x height x frame rate x bits per pixel gives the pixel payload. Blanking and protocol overhead depend on the sensor and the link, so the tool adds an overhead percentage you can edit.

Where do the numbers come from?

Each interface, sensor and stack figure links to a standards body, chip vendor, sensor maker, robot maker or test-equipment page, with the quote it was taken from. Secondary sources and derived values are labelled, and values no verified source publishes are tagged as assumptions.

Everything runs in your browser. Nothing you type is sent anywhere. See also the Robot Platform Hub and Robot Economics Calculator.