On May 28, 2026, BYD officially launched its satellite architecture sensor solution at its intelligent driving launch event, becoming the first automaker in the industry to achieve mass production and deployment of this architecture.
BYD introduced at the event that the satellite 4D millimeter-wave radar has achieved a significant enhancement in perception capabilities. The front radar can detect obstacles 400 meters in advance, with a single-frame point cloud density reaching 6,000 points, which is 6 times that of traditional solutions. The obstacle recognition rate in rain and fog has increased by nearly two times.
The day after BYD's launch event (May 29, 2026), the EAC 2026 Autonomous Driving and Embodied Intelligence Industry Exhibition concurrently held a closed-door seminar on 4D satellite architecture radar.
The meeting invited over 30 core experts from the upstream and downstream of the industry chain. Participants covered leading OEMs, leading module and chip manufacturers (company names are omitted here due to confidentiality). Attendees included corporate CEOs, CTOs, technical directors, marketing directors, and core experts in system hardware and software development.
The two-hour meeting featured in-depth discussions on the technical feasibility, cost accounting, development challenges, industrial ecosystem, and future trends of satellite architecture radar. The author hosted the meeting. This article extracts the core viewpoints and consensus of the meeting for reference by industry peers.
01. Why Adopt the Satellite Radar Architecture?
The implementation of the satellite radar architecture is not driven by a single technology, but is the result of multiple factors, including cost, performance, industry discourse power, and market competition.
The core starting point for OEMs to select components is "high quality and low price." However, against the backdrop of hyper-competition in the automotive industry, this requirement often shifts to "low price and high quality," where the weight of price sometimes outweighs performance.
1) Reducing Hardware Costs.
Traditional edge architecture radars, after years of iteration, have generally adopted highly integrated single-board designs. The structure has been streamlined to the extreme, leaving almost no room for further cost reduction.
In contrast, the satellite architecture removes the local processor at the radar end and shifts the signal processing logic to the domain controller, directly reducing the cost of a single radar by approximately 30-40 RMB.
Given the current market prices where 4T4R radars have dropped to the 90 RMB range and 8T8R radars are around 200 RMB, this cost reduction is highly attractive to cost-sensitive OEMs.
2) Enhancing Perception Performance.
The satellite architecture transmits raw ADC data from the radar directly to the domain controller. Leveraging the computing resources at the domain controller, which far exceed those of local processors, it can run high-performance algorithms such as AI (Artificial Intelligence) large models and super-resolution that traditional radars cannot support.
Combined with hardware upgrades for more antenna channels such as 8T8R and 16T16R, it can effectively address corner cases that are difficult for traditional radars to handle, such as static target recognition, near-field obstacle differentiation, and false alarms in rain and fog. This achieves higher angular resolution and lower false and missed alarm rates, truly upgrading millimeter-wave radar from an "auxiliary sensor" to a "core perception sensor."
Beyond cost and performance, there are two other auxiliary factors:
3) Enhancing OEMs' System Control.
Intelligent driving systems have become the core competitiveness of automakers. With radar algorithms shifted to the domain controller, OEMs can deeply participate in the entire process of radar signal processing and multi-sensor fusion, completely eliminating their reliance on the black-box algorithms of module manufacturers, and significantly enhancing their control over the entire intelligent driving system.
This is also one of the core reasons why the central computing architecture has become an industry trend; the strongest driving force always comes from the OEMs that control the terminal market.
4) Adding Product Marketing Selling Points.
In the current era of severe homogenization in intelligent driving technology, innovation at the architecture level can form a significant differentiation label.
As a cutting-edge technology in the industry, the satellite architecture can serve as a core highlight for brand marketing, increasing product exposure and market heat. This not only directly drives terminal sales but also sends a signal of technological leadership in the capital market, thereby boosting the company's market value.
02. Can Satellite Radar Truly Reduce Costs?
During the discussion, industry experts had certain disagreements regarding the cost-reduction effect of satellite radar.
It is understood that as early as 2023-2024, three leading German automakers, together with Bosch and Continental, completed a comprehensive evaluation of the satellite architecture. They ultimately concluded that although the solution aligns with industry development trends, it cannot achieve cost reduction at present. If it only improves performance, it lacks appeal to OEMs; therefore, related projects have been suspended.
Some domestic module manufacturers, combining their mass production practices, pointed out that the cost-reduction effect is strongly correlated with the number of radar channels and cannot be generalized:
1) Differences in Channel Numbers Determine the Cost-Reduction Effect.
The local processor cost for 4T4R radars is relatively low. Removing it yields limited savings and instead increases transmission and domain controller computing costs, making it more suitable to retain the edge architecture.
However, for high-channel radars with 8T8R and above, the cost proportion of the local processor increases significantly. The savings from removing it are sufficient to cover the new transmission and computing overhead. Moreover, the more channels there are, the greater the cost reduction—the cost reduction effect of 16T16R radars is particularly obvious.
In addition, the compression and decompression algorithms of some chip manufacturers have no practical value for the domain controller and even cause signal-to-noise ratio loss. The satellite architecture can directly remove this redundant design, further reducing costs.
2) Computing Cost Has Long-Term Optimization Space.
Relying on the more advanced process technology of central chips, the unit cost of equivalent computing power will be significantly lower than that of independent processors at the radar end, possessing significant cost-reduction potential in the long term. However, under the current architecture, the efficiency of using general-purpose GPUs to process radar signals is only 1/3 to 1/2 of that of dedicated RSPs (Radar Signal Processors). Computing power has only achieved a physical shift in location, without bringing about cost reduction.
Therefore, whether the entire system can truly achieve cost reduction in the future depends fundamentally on whether central control chip manufacturers will launch products integrating dedicated radar processing units.
3) The Impact of Other Cost Items is Negligible.
Wiring harness costs will not be an obstacle: current vehicle models generally adopt multi-domain network architectures, and the cost of Ethernet wiring harnesses has dropped significantly; switching from traditional PHY chips to higher-bandwidth SerDes chips also shows no obvious price fluctuation.
03. Does the Domain Controller Have Enough Computing Power for Radar?
The allocation of computing power for radar algorithms at the domain controller was a hot topic of discussion at this meeting.
Some chip manufacturers calculated that the signal processing of 8T8R radars requires about 8%-10% of the resources of existing intelligent driving domain controller SoCs (System on Chip). OEMs generally cannot accept this proportion of computing power occupation.
However, some module manufacturers mentioned based on practical experience that the computing power of current mainstream domain controller SoCs actually has a certain amount of idle capacity. Radar algorithms can be satisfied with only a small amount of computing resources, without occupying core computing units such as NPU, GPU, and CPU, and DDR bandwidth occupation is also low. The main challenge lies in the allocation of low-latency data bandwidth, which needs to be solved through engineering capabilities.
From another perspective, the core contradiction in computing power allocation is not whether the absolute computing power is sufficient, but whether the computing power investment matches the performance benefits.
The premise for OEMs to be willing to pay for computing power is: "Can a 10% investment in computing power bring a 20% or even higher performance improvement?" However, most radar manufacturers cannot provide quantitative data on the improvement of false and missed alarm rates under a centralized architecture, making it difficult to prove the value of radar at the system level.
From the perspective of chip manufacturers, one of the biggest core competencies comes from the design of the dedicated radar signal processor (RSP), which often directly determines the core performance of the entire radar.
It is understood that domestic domain controller manufacturers are already deploying RSP IP. Once a technological breakthrough is achieved, the central integration of millimeter-wave radar will accelerate its deployment. When the number of radars installed per vehicle exceeds 5, it will also further promote the rapid development of RSPs.
04. What Are the Development Challenges of Satellite Radar?
1) High System Development Difficulty
Some experts mentioned that there are few domestic manufacturers proficient in radar module and domain controller integration technologies. There are numerous issues with communication handshakes and interface compatibility between domain controllers and radar modules.
In addition, signal integrity issues in vehicle environments have always been a challenge. It often happens that suppliers test modules, wiring harnesses, and domain controllers individually with normal results, but failures occur after installation in the vehicle. The raw data transmitted by the satellite architecture belongs to high-speed signals, and these issues exist equally. Domestic OEMs generally have weak capabilities in high-frequency signal transmission.
Meanwhile, cameras only need to passively transmit data, whereas radars require active control. Therefore, radars are highly likely to encounter all the interaction problems that cameras face, and the problems will be more complex.
2) Increased Software Complexity
After the satellite architecture shifts all radar signal processing logic to the domain controller, the software complexity of the central controller increases significantly, the probability of failures rises accordingly, and the difficulty and cost of later software maintenance will also increase substantially.
3) Greater Functional Safety Challenges
Under the traditional edge architecture, radars can directly output braking/acceleration signals and possess independent ASIL (Automotive Safety Integrity Level) ratings. Under the satellite architecture, all commands must be forwarded by the central controller, changing the system redundancy, and functional safety designs need to be readjusted.
If deep learning end-to-end algorithms are adopted, bypassing the traditional signal processing flow, it will form an algorithm "black box," making fault tracing difficult. During the discussion, industry experts generally believed that visual solutions could be referenced and solved through system-level redundancy, intermediate result verification, and failure mechanism design.
4) Introduction of Cybersecurity Issues
Currently, the raw signals of satellite architecture radars are transmitted in a bare state, making them highly susceptible to tampering by cyberattacks. As the safety fallback sensor for intelligent driving systems, once attacked, radars will directly threaten the overall vehicle safety. In the future, the camera CSE v2 encryption scheme can be referenced to design dedicated radar transmission encryption chips, realizing the encryption and signing of raw data.
05. Ecosystem
1) Will OEMs Develop Software Themselves?
Some experts believe that the collaboration between OEMs and module manufacturers will be progressively advanced in three stages:
Stage 1: Module manufacturers provide complete radar products and algorithm SDKs, and OEMs are only responsible for integrated applications.
Stage 2: As industry competition intensifies, module manufacturers open white boxes to OEMs, and OEMs gradually learn and master radar algorithms and system design.
Stage 3: OEMs fully master radar technology, independently maintain radar algorithms, and integrate them into end-to-end intelligent driving systems.
However, some experts also believe that radar software development requires a massive investment of personnel, and most OEMs may not fully undertake radar software development.
Furthermore, radar is a highly non-standard product. There are no unified industry standards for waveform settings, antenna array layouts, and algorithms. This not only increases the difficulty of in-house development for OEMs but also means that the future industrial ecosystem may rely more on deep cooperation between upstream and downstream.
Therefore, the collaboration model between OEMs and module manufacturers needs further observation.
2) The Dilemma of Module Manufacturers?
The industry is currently suffering from severe low-level hyper-competition. Module manufacturers generally lack R&D funding, investment in cutting-edge technology research continues to decline, and effective innovation outcomes are scarce.
During the meeting, a module manufacturer shared its current survival status, reflecting the typical hyper-competition dilemma in the industry and resonating with many industry peers.
No recognition: They undertake all the dirty and tiring work and risks of R&D and mass production, but after mass production, they cannot gain brand exposure, and cannot even publicly claim to be the supplier of the product.
Unprofitable: Even with cumulative shipments of tens of millions of units, profits remain meager. Huge investments are required to build factories to ensure supply, and profit margins are further compressed after capacity ramp-up.
Little forward-looking research: Meager profits cannot support cutting-edge technology R&D, falling into a vicious cycle of "low-price competition - no R&D investment - product homogenization - lower-price competition."
Risk of customer in-house development: After module manufacturers help customers complete technical verification and capacity construction, customers turn to in-house development, leaving module manufacturers facing the risk of order loss and previous production line investments.
Some experts called for promoting joint R&D of cutting-edge technologies through industry associations and social organizations, striving for support from major national projects, concentrating efforts to break through core technologies such as radar algorithm large models and distributed coherent radar, avoiding homogenized competition, and leaving room for innovation and development for module manufacturers.
3) Can Radar Break Through in Other Fields?
Beyond the automotive field, industry experts generally suggest that radar manufacturers expand into non-automotive fields to find new growth points, such as:
Verified high-value fields: Radar applications in construction machinery, special vehicles, and low-speed vehicles have achieved profitability, with a global market scale reaching tens of millions of units.
Rapidly growing emerging fields: The explosion of the low-altitude economy has driven the demand for low-altitude radar. The cost of traditional low-altitude radar is as high as hundreds of thousands of RMB. After technological transformation, automotive radar has a significant cost advantage, with huge future potential.
Livelihood segmented fields: Segmented markets such as human presence monitoring in smart homes and fall detection in the elderly care field are huge and have not been fully explored.
However, some manufacturers also stated that although the unit price in non-automotive fields is high, the market distance is relatively far, and some fields also have hyper-competition issues, making the transformation path unclear.
06. Outlook
1) The Architecture May Not Be the Final Outcome
Technical architectures are not static and will be dynamically adjusted according to actual needs. Referring to the development history of mobile phone GPS: from a complete functional module to retaining only the satellite front end, and then returning to the edge module due to low-power requirements.
In the future, if requirements such as low power consumption and real-time performance change, the possibility of some radar functions returning to the edge side cannot be ruled out.
2) Software Architecture Transformation
The current satellite architecture only migrates algorithms from the radar end to the domain controller. The algorithms themselves have not undergone essential changes, and the practical significance is limited.
In the future, it is necessary to develop deep learning end-to-end algorithms and multi-sensor low-level fusion algorithms based on raw data, using AI data-driven approaches to improve radar performance. Only then can the value of the satellite architecture be truly released, avoiding radar being relegated to a supporting role in the intelligent driving system forever.
3) Distributed Radar
Some experts believe that satellite architecture radar is only the first step in radar technology development, and distributed coherent radar is the future. If RF phase coherence or IF phase coherence can be achieved, radar performance will achieve explosive improvement, completely changing its supporting role in the intelligent driving system.
However, the engineering difficulty of this technology is extremely high. It is still in the early stages, and its deployment requires solving a large number of vehicle-side engineering problems.
07. Conclusion
Over the past decade, automotive millimeter-wave radar has continuously achieved performance improvements, cost reductions, and increased penetration rates. However, the development of the entire industry has also fallen into a vicious circle.
Throughout the meeting discussion, the author deeply felt the anxiety and even helplessness of many millimeter-wave radar practitioners. However, from everyone's active discussion, one can equally feel their passion for millimeter-wave radar and their deep expectations for the future.
During the meeting, participants generally recognized the development direction of the satellite architecture. Like the participating experts, the author also expects the satellite radar solution to break the current vicious circle and promote the industry to form a positive cycle of "high performance - high value - high profit - high R&D investment."
In the process of this positive cycle, participants at every link in the industry chain can find a more suitable positioning for themselves.
The author will continue to organize satellite radar-related seminars in the future, and related information will be synchronized in the millimeter-wave radar expert community.
Since the communities all exceed 200 members, teachers interested are welcome to add the author's WeChat (scan the QR code at the end of the article or search maxhnnl) for an invitation to join.
Personal viewpoints, not necessarily accurate, discussions are welcome.
I am Xueling, researching the technologies, products, and applications of perception, control, and AI. Welcome to communicate.