4D Millimeter Wave Radar

Millimeter wave radar in smart driving sensors is good at measuring distance and speed, and has good distance and speed resolution. Compared with other sensors, radar operates at night and has minimal impact in bad weather conditions.

It is also claimed that radar is the second cheapest sensor on a car after ultrasonic sensors. Based on these advantages, the technological development and application of radar in intelligent driving are constantly advancing.

For example, the article "4D Imaging Millimeter-Wave Radar - 101" states that the current direction of millimeter-wave radar is to improve the angular resolution to become a 4D millimeter-wave radar, and to continue to advance the provision of data close to the camera image, called 4D millimeter-wave imaging radar, which can realize the application of AI algorithms like the vision "Vision is King - Xiaopeng and Tesla's Autonomous Driving Solutions".

At present, the technical path of 4D millimeter wave is mainly to increase the number of antenna channels, that is, the number of receiving and transmitting, to achieve richer perception data. So in theory, the more the perceived receiving and transmitting data, the richer and more accurate the data, and of course the greater the amount of information processing.

Mobileye claims that in the future, radars with thousands of channels can be used for point cloud and other AI algorithms, and can identify more than a thousand targets. Of course, it is difficult for it to replace LiDAR. After all, due to the length of the wave it uses, the wavelength is long, the resolution is low, and the edge information is definitely not as rich and accurate as that of LiDAR.

Therefore, the current innovation of 4D millimeter-wave radar is basically focused on how to have more perception and communication channels and how to process these signals to obtain the desired information. The current mainstream technology chain is as follows:

• Standard radar chips (NXP, Texas Instruments, etc.) are cascaded to increase the number of antennas, and 4D effects are achieved through software processing innovations. Traditional giants such as Continental and Bosch, as well as domestic companies such as Huawei, integrate multiple transmit and receive antennas into one chip, and directly provide imaging radar chips, such as Arbe, Vayyar, mobileye, and uhnder. The most difficult thing is to develop a new radar architecture antenna array through metamaterials, and representative manufacturers include Metawave, etc.

Standard radar chips are cascaded to increase the number of antennas. This is the current mainstream traditional solution, which basically comes from the technical solutions of the following three suppliers:

• 
  

• Based on the front-end transceiver of Texas Instruments AWR series chips, using TDA2X series chips for processing. NXP's TEF series chips for front-end transceiver, S32R series chips for processing. Xilinx FPGA chip solution.

Use 2-transmit/4-receive or 3-transmit/4-receive, about 45nm chip basic solution for cascading and increasing channels.

Currently, such chip companies are building a development ecosystem. The underlying software development is very complete, and the tool chain is easy to use, making it easy for more people to use and develop. For example, as mentioned earlier in "Nvidia's Belief in AI Smart Cars", the millimeter-wave radar chips of these chip suppliers are no exception. This has spawned a large number of domestic start-up millimeter-wave radar companies. It is rumored that Huawei's radar solution also uses the above solution. If this is the case, the cold winter that Mr. Ren mentioned is coming, and these businesses should be eliminated. Not to mention whether they can make money, even if they make money, their life will be in the hands of others, just like mobile phones. Why bother?

Of course, the traditional T1 companies such as Bosch, Continental, and ZF do not just use chips. Their know-how in the automotive industry enables them to make precise demands on radar chips, which also promotes the development of the chip industry. In addition, there is the innovative force Oculii, which is backed by Ambarella chips. It uses SLAM-style mechanisms to process acquired data based on chip solutions from Texas Instruments and Infineon.

The goal is to compare multiple radar frames to build a more detailed and accurate picture of the environment than any single frame. Oculii's software also uses ML machine learning/DL deep learning technology to adjust the transmission signal parameters to adapt to specific road environments, claiming that its algorithm can achieve a maximum performance improvement of 100 times. In addition, Infineon, the protagonist of traditional radar, is also making efforts to catch up. Integrating multiple transmitting and receiving antennas on a chip is basically a chip design and innovative technology. At present, the innovative new forces of 4D millimeter wave radar are concentrated in this part.

Arbe's transceiver currently has a 24TX/12RX channel configuration. It requires the use of Global Foundries' new FDSOI CMOS 22FDX technology, which is more complex than some CPUs using 22nm technology. Arbe also built a processing chip that can manage up to 48 Rx channels and 48 Tx channels in real time, generating full 4D images at 30 frames per second, with an equivalent processing throughput of 3 Tb/sec.

Another silicon chip prodigy, Vayyar's has 72 transmit channels and 72 receive channels, forming more than 2,000 virtual channels. It is a true imaging radar that can generate high-resolution 3D images of objects. When there are so many channels, power consumption becomes its limitation. Developers must provide acceptable isolation levels for so many channels by limiting the peak output power of the radar, etc. Therefore, Vayyar focuses on in-vehicle monitoring applications and short-range radar cases.

Uhnder, whose product is a 28 nm, 12TX/16RX channel transceiver with CPU, DSP, memory and interfaces on a monolithic silicon die. These numbers should not be directly compared to competitors. Uhnder uses its own software to implement phase modulated continuous waveform (PMCW) with digital coded modulation (DCM) instead of the more common frequency modulated continuous waveform (FMCW) approach. It also helps eliminate mutual radar interference by using a nearly unique phase coded detection signal.

Mobileye, backed by Intel's chip technology, is equipped with 2,304 virtual channels based on 48 x 48 transmission and reception, and realizes perception based on frequency modulated continuous wave (FMCW) technology and Doppler algorithm. Therefore, millimeter wave radar has also embarked on the road of intelligent AI volume parameters similar to cameras and lidar, 4D millimeter wave volume channel number, volume signal processing, volume algorithm and processing unit capabilities, intelligent AI road. But in fact, in the world of intelligent AI applications, perception channels, signal processing, and algorithm processing are all based on cost, energy consumption, and object volume. Artistic applications need to be balanced. For example, if the perception is strong enough, it is also a waste if the algorithm and data standards cannot keep up. Therefore, as developers and consumers, there is no need to pursue strong parameters, but also to look at the efficacy. At present, the pioneer of 4D millimeter wave is BMW's IX. If you have information and know its performance, you can send me a private message to see if you can analyze the connection between parameters and performance.