Perception & Sensing

Source discipline

The strongest sources used in this note:

The technical performance claims below rest primarily on S1, S2, S3, S4, and S8. Company disclosures in S5–S7 are used for commercialization context, not for physical performance proof.

Technology fundamentals: what “perception-and-sensing” means here

Perception-and-sensing converts an environment into structured, machine-consumable data. Cameras capture reflected visible light; radar captures RF reflectivity and Doppler; LiDAR captures active near-infrared or short-wave-infrared time-of-flight or Doppler range data. LiDAR’s key advantage is direct geometric range output, but its cost, eye-safety limits, and weather sensitivity remain contested S2.

The two dominant LiDAR architectures relevant to current public-company claims are:

1. Direct time-of-flight (dToF). A laser pulse is emitted, a detector records return time \( \Delta t \), and range is:

\[

R = \frac{c \Delta t}{2}

\]

where \( c \) is the speed of light. The factor of 2 accounts for the round trip. This is the architecture at issue in the Velodyne granted patent discussed below S1.

1. Frequency-modulated continuous-wave (FMCW). A frequency sweep is transmitted and mixed with the returning signal. Range is recovered from the beat frequency, and radial velocity is recovered from Doppler shift:

\[

f_D = \frac{2 v}{\lambda}

\]

For a target moving at \( v = 1\ \text{m/s} \) at \( \lambda = 1550\ \text{nm} \):

\[

f_D = \frac{2 \times 1}{1550 \times 10^{-9}} \approx 1.29 \times 10^6\ \text{Hz} = 1.29\ \text{MHz}

\]

That means coherent FMCW can measure velocity directly per frame, while dToF systems must infer velocity from multiple frames S2. This is not a minor software detail; it is a physical difference in signal architecture.

Granted patent analysis: US 7,969,558 B2

The granted patent S1 is titled “High Definition LiDAR System”, assigned to Velodyne Acoustics, Inc., and was granted on 2011-07-28. It is a granted patent, not an application. That distinction matters: a granted patent has survived examination and is enforceable until term limits, but it does not certify that the disclosed device achieved any specific performance or commercial success.

The specification describes a rotating housing carrying a plurality of laser emitters and detectors arranged to cover vertical field-of-view sectors. In one embodiment, the patent describes 64 laser/detector pairs, which corresponds to the architecture publicly associated with Velodyne’s HDL-64E class of mechanically rotating LiDAR S1. The claims are structural: they protect the scanner arrangement, not a specific range algorithm or semiconductor process.

Moat interpretation. This patent supports the historical claim that Velodyne held an early mechanical-scanning moat. But the commercial result is a disconfirming case for “one patent = durable moat.” Velodyne later merged with Ouster; Ouster’s SEC filings continue to describe a business that must execute on digital LiDAR, manufacturing scale, and customer concentration risk S7. The patent did not prevent later solid-state, MEMS, and FMCW architectures from emerging.

Technical cross-check. The IEEE paper S2 describes mechanical rotating LiDAR as mature but physically limited in size, cost, and angular resolution. The fact that a peer-reviewed survey paper can describe the same class of device does not independently validate Velodyne’s commercial claims; it validates that rotating dToF LiDAR is a known, reproducible technical category.

Peer-reviewed paper analysis: IEEE Signal Processing Magazine, 2020

The IEEE paper by Li and Ibanez-Guzman, “LiDAR for Autonomous Driving: The Principles, Challenges, and Trends for Automotive LiDAR and Perception Systems”, was published in IEEE Signal Processing Magazine in July 2020 with DOI 10.1109/MSP.2020.2973615 S2. Unlike an arXiv preprint, this is a journal-published, DOI-verified Tier 1 source.

The paper’s relevant technical claims:

• LiDAR performance is a multi-parameter trade among range, field of view, angular resolution, frame rate, eye safety, cost, and mechanical reliability S2.
• Mechanical rotating LiDAR achieved early success but has disadvantages in size and long-term reliability S2.
• 905 nm systems benefit from silicon photodetector availability; 1550 nm enables higher eye-safety thresholds but historically required more expensive non-silicon detectors or optics S2.
• Automotive LiDAR must operate under sunlight, rain, snow, fog, vehicle vibration, and thermal extremes; any single lab-range spec is insufficient on its own S2.

This source does not endorse any specific company. It is useful as an independent baseline for what the upstream physics demands.

Standards and eye-safety constraint

Automotive perception sensors must satisfy functional safety and laser safety frameworks.

ISO 26262. The functional safety standard ISO 26262-1:2018 covers road-vehicle electrical/electronic systems S3. A LiDAR vendor’s claim of being “automotive qualified” should be read against this standard, not merely against a datasheet number.

IEC 60825-1. Laser product safety is governed by IEC 60825-1:2014 S4. This is a Tier 1 standards-body document and the source for the claim that laser exposure limits depend on wavelength. At 1550 nm, the eye’s anterior segment absorbs light before it reaches the retina, allowing higher emitted power within a given eye-safety classification than 905 nm. This is a load-bearing physical/regulatory claim and is supported by the IEEE paper S2 plus the laser-safety standard S4 — two independent lineages, satisfying the two-source rule for this claim.

Commercial evidence and 10-K cross-reference

Layer 1: Marketing/product evidence. Luminar’s newsroom release dated 2022-11-09 announces that Volvo Cars revealed the EX90 with Luminar LiDAR as standard safety equipment S5. This is a named customer/product-release datum, but it is Tier 2 self-reported. It proves an announced product integration, not annual volume, margin, or profitability.

Layer 2: Filed disclosures. Luminar’s SEC EDGAR 10-K index, CIK 0001810930, contains repeated risk factors that the company has a history of losses and expects to continue incurring substantial losses; that automotive OEM adoption may not reach mass production; and that supply-chain or customer concentration may harm operations S6. This is not a contradiction of the Volvo announcement in the sense of factual falsification. It is a contradiction in implied confidence: a product announcement is not the same as a contracted, scaled, profitable product launch.

Layer 3: Continuation-of-commerce evidence. Ouster’s EDGAR filing index, CIK 0001816581, reflects a public company that must manufacture, sell, and support digital LiDAR units while managing going-concern and/or liquidity risk typical of pre-profitability hardware businesses S7. The existence of these risk factors does not mean Ouster cannot succeed; it means the investor-facing moat claim cannot rest only on patent count or tech marketing.

Worked quantitative calculation

7.1 dToF round-trip range resolution

For a pulse width \( \tau \), the basic two-way range-resolution limit is:

\[

\Delta R = \frac{c \tau}{2}

\]

Using a representative automotive dToF pulse width of 5 ns:

\[

\Delta R = \frac{2.99792458 \times 10^8 \ \text{m/s} \times 5 \times 10^{-9} \ \text{s}}{2}

\]

\[

\Delta R \approx 0.75\ \text{m}

\]

This is a simplified physical lower bound on range separation between two objects at the same angular cell when using pulse-width-limited detection. Timing interpolation or multi-echo processing can improve nominal resolution, but the pulse-width floor remains a useful engineering benchmark. The speed-of-light value is from NIST CODATA S8; the dToF round-trip formula is standard and appears in the IEEE paper’s treatment of direct time-of-flight S2.

7.2 Photon budget at 200 m for 10% Lambertian target

We estimate received optical power for a small-aperture LiDAR receiver under the simplifying assumption that the target is larger than the beam, fully illuminated, and the receiver field of view covers the target:

\[

P_r = P_t \cdot \rho \cdot \frac{A_r}{\pi R^2} \cdot \eta_{\text{total}}

\]

Inputs:

• \( P_t = 100\ \text{W} \) peak transmit power — assumed for illustration, not a company-verified spec.
• \( \rho = 0.10 \) Lambertian reflectance — assumed common lidar reflectance benchmark.
• \( A_r = \pi (0.010\ \text{m})^2 = 3.14159 \times 10^{-4}\ \text{m}^2 \) for a 20 mm receiver aperture — assumed.
• \( R = 200\ \text{m} \).
• \( \eta_{\text{total}} = 0.49 \) combined optics efficiency — assumed.

\[

\pi R^2 = \pi (200)^2 = 125{,}663.7\ \text{m}^2

\]

\[

P_r = 100 \cdot 0.10 \cdot \frac{3.14159 \times 10^{-4}}{125{,}663.7} \cdot 0.49

\]

\[

P_r \approx 1.23 \times 10^{-8}\ \text{W} = 12.3\ \text{nW}

\]

Photon energy at \( \lambda = 905\ \text{nm} \):

\[

E_{\text{ph}} = \frac{hc}{\lambda}

\]

Using \( h = 6.62607015 \times 10^{-34}\ \text{J·s} \) and \( c = 2.99792458 \times 10^8\ \text{m/s} \) S8:

\[

E_{\text{ph}} \approx \frac{1.98645 \times 10^{-25}}{905 \times 10^{-9}} \approx 2.195 \times 10^{-19}\ \text{J}

\]

Photon arrival rate:

\[

N_{\text{rate}} = \frac{P_r}{E_{\text{ph}}} \approx \frac{1.23 \times 10^{-8}}{2.195 \times 10^{-19}} \approx 5.6 \times 10^{10}\ \text{photons/s}

\]

For a 5 ns pulse:

\[

N_{\text{pulse}} = 5.6 \times 10^{10} \times 5 \times 10^{-9} \approx 280\ \text{photons}

\]

If the detector has a photon detection efficiency \( PDE = 0.20 \), detected photoelectrons per pulse are about:

\[

280 \times 0.20 \approx 56

\]

Sensitivity. If reflectance drops from 10% to 1%, the signal drops linearly to about 28 photons, or about 5.6 detected photoelectrons per pulse. If range doubles from 200 m to 400 m, received power drops by \( 1/R^2 \) to one-fourth, yielding about 70 photons or about 14 detected photoelectrons at the same PDE. If aperture diameter doubles from 20 mm to 40 mm, collection area rises by 4×, yielding about 1,120 photons per pulse. This is the central physical constraint on automotive LiDAR: range, aperture, pulse energy, and target reflectance are coupled.

This calculation is deliberately not a company performance claim. It is a first-order physical model using standard constants and assumed engineering inputs.

7.3 Parameter table

No cell in this table should be read as a company-specified sensor capability.

Contradictions and source-bound claims

Contradiction 1: patent category vs. market movement

The granted patent S1 protects a rotating multi-channel dToF architecture. The peer-reviewed survey S2 reports that mechanical LiDAR is mature but faces size, reliability, and cost pressure from solid-state and FMCW alternatives. A legal asset from 2011 therefore does not automatically map to the 2024–2026 technology frontier.

Contradiction 2: product announcement vs. filed risk factors

Luminar’s newsroom announcement says the Volvo EX90 reveal included Luminar LiDAR as standard S5. Luminar’s SEC EDGAR filing index shows the company’s risk disclosures include continued losses, adoption uncertainty, and customer concentration S6. The reconciliation is not fraud; it is that a reveal is an engineering/program milestone, not a unit-economics milestone.

Contradiction 3: “high definition” claim vs. physics

A patent title saying “high definition” S1 is a legal/descriptive claim. But angular resolution in rotating systems is limited by mechanical indexing, channel count, and point rate S2. This is an example of why patent language is not a performance benchmark.

Unit economics / business model

Who monetizes. LiDAR suppliers monetize by selling sensor hardware, charging non-recurring engineering for OEM integration, licensing perception software, and sometimes selling reference systems or data-annotation/development kits. The public-company record through EDGAR shows that these businesses are hardware-heavy, pre-scale, and mostly loss-making at the operating line S6 S7.

Margin reality. Specific current gross-margin percentages for individual LiDAR pure plays are not stable enough to cite as a durable fact in this note without primary source validation; any such figure would be labeled needs primary source validation. What can be cited from filings is that the companies disclose substantial historical losses and do not assure profitability S6 S7.

Value capture. The value chain divides roughly into:

• Laser/detector suppliers capture component revenue: semiconductor lasers, VCSELs, SPAD/SiPM arrays, and control ASICs.
• Optics and MEMS suppliers capture precision-optic and scanner revenue.
• LiDAR system suppliers capture sensor assembly and embedded perception revenue.
• Tier-1/OEM integrators capture vehicle integration and safety-validation value.
• OEM consumers capture the end-product value only if the feature drives pricing, insurance, regulatory compliance, or autonomy service monetization.

This is not a “software-like gross margin” story at the system level. The hardware cost and automotive qualification burden suppress the margin profile relative to pure perception-software businesses.

Disconfirming Evidence

The thesis that automotive LiDAR is a high-value, patent-backed perception-and-sensing weapon would be wrong if:

1. Mass automotive adoption does not occur. If Luminar, Ouster, Innoviz, Aeva, Hesai, and MicroVision continue to report negative operating income and cannot convert named OEM programs into growing sensor shipments, the component and system value thesis fails S6 S7.
2. Camera or radar perception closes the performance gap. If camera-only or camera-plus-radar systems achieve sufficient depth and object-detection performance at lower cost, LiDAR’s range advantage may not justify its price, assembly complexity, and supply-chain burden S2.
3. Price compression eliminates system-level margins. If Chinese suppliers and vertically integrated tier-1s push LiDAR unit prices below the cost structure of Western pure plays, hardware suppliers may never reach sustained gross profit.
4. Regulation does not mandate LiDAR. If safety regulators do not require active LiDAR-based performance and insurers do not differentiate it, the demand pull may be too weak to support multiple public-company cost structures.
5. A single architecture or supplier wins completely. The current architecture race between dToF, FMCW, and future solid-state designs S2 means that any granted patent may become a cost, not a moat, if the industry moves to a different scanner or coherent-detection architecture.