Microwave Journal
www.microwavejournal.com/articles/38807-advanced-multi-mode-multi-mission-software-defined-mmwave-radar-for-low-size-weight-power-and-cost

Advanced Multi-Mode Multi-Mission Software-Defined mmWave Radar for Low Size, Weight, Power and Cost

September 8, 2022

A series of advanced electronically scanned phased array (AESA) mmWave radars are designed with a multi-mode multi-mission software-defined radar (SDR) capability. These research radars address a variety of markets including advanced driver assistance systems (ADAS), fixed or mobile ground deployed small unmanned aerial vehicles (sUAVs) drone detection and tracking systems, sUAV air-to-air and air-to-ground radars and sUAV deployed airborne synthetic aperture radar (SAR). The radars are designed to facilitate radar research and development from early stage concept-of-operations through requirements definition and validation to system design, verification and deployment.

The radars are designed and manufactured by aiRadar Inc., using highly integrated, state-of-the-art RFICs from Sivers Semiconductors AB for the transmit/receive modules (TRMs). All are multi-mode multi-mission. They can switch seamlessly between a sector scanner covering 90 degrees in azimuth with better than 0.5 degrees angular resolution with triple baseline interferometric positioning in elevation, to a sUAV deployed single pass interferometric SAR (InSAR) with range and azimuth resolution better than 5 cm generating digital surface models (DSMs) with up to 16 channels of along track interferometry for high-resolution unambiguous moving target indication (MTI).

The InSAR configuration provides a multi-aperture SAR capability with displaced phase center antenna (DPCA) micronavigation. The radars range in size from the smallest model, with a mass of 3,850 grams comprising three transmit (Tx) and three receive (Rx) 64-element arrays, to the largest model, with a mass of less than 10 kg comprising 1,536 active elements in an identical array layout but with 256-element arrays.

The target customers are commercial, military and academic researchers who seek to advance the state-of-the-art in radar using a ruggedized reconfigurable instrument rated to IP-67 and Mil-Std-810. Deploying these research radars with a simple but powerful compiled radar programming language (aiRPL), which executes on a field-programmable gate array based multi-mode radar processing unit (aiRPU), eliminates the risks associated with developing AESA radar systems based on analysis and simulation, either computer simulation or over-the-air validation using radar target simulators.

An ADAS developer might wish to perform an operational real-world comparison of various AESA configurations with different array sizes and with virtual or real array elements. These research radars permit, for example, the direct co-located and co-temporal comparison of a two Tx and four Rx MIMO array to a 12 Tx and 16 Rx MIMO array. Similarly, the 12 Tx and 16 Rx MIMO array (with 192 virtual channels) may be compared to a 256 Tx and 512 Rx array (with 512 real channels arranged as a long or medium baseline elevation interferometer). A simple script in aiRPL manages the complexity, enabling these three (or more radar configurations) to be cycled on a PRI-by-PRI basis, providing an objective comparison of radar performance under the same operating conditions.

Once the requirements and AESA configurations are validated for a specific application, the commercial, military or academic radar developer may proceed, based on the risk assessment, the economics or the urgency of time-to-market, with an in-house radar design or an aiRadar customized application specific radar. This can be done with or without the licensing of the aiRadar programming language compiler and the radar processing unit IP Core.

The aiRPU IP Core provides real-time bidirectional interfaces, up to 48 Gbps, to the lowest level in-phase and quadrature raw radar data channels, and to the aiRPU IP Core. This interface is provided for researchers and developers of cognitive adaptive (CA) radar allowing an external artificial intelligence (AI) processor, perhaps based on GPU arrays, to modify any or all of the radar configurations from transmit pulses to beamforming/steering directions on a PRI-to-PRI bases.

An example application is adaptive pulse code modulation (PCM) for ADAS in the inevitably congested radar environment that will exist as more radars are deployed in ever more advanced systems. The CA loop facilitates the analysis of received signals to determine if an interfering source (another vehicle) is present and select the PCM codes to reject that interference. This CA loop has applications in low probability of intercept (LPI) radars for military applications as well. A key feature of the CA physical and API interfaces is that the algorithms in the CA loop remain the exclusive intellectual property of their developers.

To facilitate the granting of experimental and research licenses, the first offering of these research radars has a center frequency at 66 GHz, where there is little commercial activity at this stage. The research radars are architected in such a manner that the digital control and RF interfaces to the TRMs enable hardware reconfiguration to 24 GHz with existing Sivers Semiconductors technology or reconfiguration to 76 to 81 GHz with future Sivers Semiconductors technology. The generalized TRM interfaces anticipate the emergence of new allocated mmWave frequency bands, should they arise.

ADAS

Given the ability of radar sensors to operate in conditions such as rain, fog and snow, which impairs or disables the operation of LiDAR sensors and visual cameras, it is inevitable that radar will become a fundamental element of ADAS.

Most radars currently deployed in automotive applications have very coarse resolution. While lower resolution radars may detect an object, a motorcycle, person or a truck, the object is represented by little more than a “blob.” The task of object recognition is largely offloaded to an AI/machine earning (ML) algorithm, where advances in AI hardware and software algorithms are tasked with providing that one crucial step closer to a fully autonomous, safe vehicle.

There may be several reasons for this allocation of functionality and performance, but one likely contender may be that the requirements definition and validation, followed by the design and manufacture of advanced modern radar with complex AESA antennas is difficult. This difficulty translates into technical, performance, schedule and cost risk. Availability of low-cost and low risk advanced AESAs may enable changes to this allocation and, perhaps, advances in autonomy levels.

In addition to the design and manufacture of complex advanced AESA radars, verification and ongoing product assurance is not trivial and requires well-defined metrics. A simple requirement such as integrated sidelobe ratio (ISLR) impacts the angular resolution of two targets and the angular measurement accuracy of a single target, as well as having a significant impact on image quality. This lack of resolution and image quality may have a very negative impact on the AI/ML interpretation of the scene.

MILITARY AND COMMERCIAL RADAR SYSTEMS

A growing number of radar applications have emerged recently where current radars perform poorly or are not suitable. These applications include ground deployed (and human portable) real aperture radar for detecting and monitoring small drones which pose security and military threats as well as small UAV-deployed high-resolution imaging with SAR and/or real aperture radar (RAR).

A good example is InSAR deployed at the site of a flooding disaster where the desired product is high-resolution scenes superimposed on DSMs, captured in real time as riverbanks and slopes subside, with the identification of objects of interest with overlays of velocity vectors (MTI) attached to those objects.

Military applications are highlighted by the ongoing conflict in the Ukraine. Hostile drones are extremely dangerous: locating forces, providing intel on those forces, directing artillery fire with devastating accuracy and assessing damage. While the open literature shows many sUAV detection systems, these do not appear to have been effectively deployed as evidenced by multiple videos from loitering drones eliciting no evasive responses when artillery is spotted onto a target or when improvised weapons, such as modified rocket propelled grenades are dropped vertically onto targets from sUAVs.

SAR offers extremely high-resolution but requires motion, while RAR provides excellent image quality from a stationary position. A sUAV deployed SAR with 3D InSAR DSMs might be the preferred instrument for pre and post operation high-resolution threat and damage assessment, while an sUAV deployed RAR with an AESA may be better suited for real-time target spotting.

ARRAY ARCHITECTURE AND TRANSMIT RECEIVE MODULES

Addressing multiple applications, multiple modes and multiple missions with a SDR on a single hardware platform with a common interface as an economic and affordable solution is challenging. An early decision to implement a hybrid beamforming architecture, with analog beamforming at the level of 16 antenna elements and digital beamforming at a higher level, reduced the number of ADCs and data rates.

An investigation of available multi-channel mmWave technologies suitable for this hybrid architecture led to Sivers Semiconductors AB. Sivers Semiconductors develops, among other things, MMICs, modules and subsystems based on advanced semiconductor technology for WiGig mmWave networks.



The Sivers TRXBF01 RFIC is integrated into a module with a 16-element Tx and 16-element Rx arrays that covers 14 GHz of bandwidth from 57 to 71 GHz. The Sivers module has a transmit power of +11 dBm per channel and a receive noise figure of 7 dB in a 90-degree horizontally scanned AESA. Figure 1 shows the front of the Sivers BFM01 module. These RFIC modules are supported by evaluation kits.

Figure 1

Figure 1 Antenna side of the Sivers Semiconductors transceiver.

Figure 2

Figure 2 RRI-100 research radar interferometer.


Figure 3

Figure 3 RRI-400 research radar interferometer.

A customized version with interfaces for coherent multi-module AESAs with wide bandwidth modulation has been developed specifically for aiRadar. This device, the BFM06012-RFM, has a modulation input with 4 GHz of transmit bandwidth, enabling 5 cm range resolution. The vertical beamwidth is modified with tapering to produce a 30-degree beamwidth with sidelobe levels below 20 dB. aiRadar has integrated these modules into research radars, the smaller RRI-100 is shown in Figure 2, the larger RRI-400 in Figure 3.

In both figures, the Tx/Rx row pairs are visible in the radome window recesses, with a pair at the top of the radome, a pair in the middle and a pair at the bottom. The spacing from top to middle is slightly different from the middle to bottom spacing. A double difference interferogram may be formed resulting in a virtual short baseline interferometer.

The aiRadar sensor electronics modules (SEMs) are configurable in frequency. A SEM can be mechanically modified to use a customized version of the Sivers Semiconductors BFM02801 for operation in the 24 GHz band, albeit with reduced bandwidth to comply with the frequency allocation. aiRadar and Sivers are currently evaluating a 77 to 81 GHz band TRM for migration from a research 66 GHz ADAS to an operational 77 to 81 GHz version.

Figure 4

Figure 4 Transceiver RFIC architecture.

Figure 5

Figure 5 Software-defined radar block diagram.

Figure 4 shows the internal architecture of the Sivers TRM, and Figure 5 shows a functional block diagram of the radar. The digitized raw Rx outputs from the radar that appear in the radar data packets may be configured (in the RRI-400) as 16 digitized in-phase and 16 quadrature (I/Q) channels from each of the three azimuth Rx arrays, or a single digitally beamformed receive signal from each of the Rx arrays.

The 16 channels of Rx I/Q data provide the multi-aperture (16) SAR capability for 5 cm along track strip map imaging, the multi-baseline MTI capability using along track interferometry and the data for the DPCA micronavigation system.

The Sivers module provides zero IF Rx bandwidth to the aiRadar SEM with multiple control interfaces: a general purpose interface (GPIO), a serial programmable interface and a beamforming control interface. The aiRadar customized Sivers module has an external 22 GHz local oscillator (LO) interface with an internal 3× multiplier to the 4 GHz wide 66 GHz transmitter.

The Sivers RF LO interface, with a 1.33 GHz bandwidth at 22 GHz, and the Tx IQ interface are driven from the dual direct digital synthesizer (DDS) in the SEM, with modulation at two levels, on both the LO and the Tx IQ. This interface providers linear frequency modulated (FMCW) and arbitrary pulse modulation supporting LPI operations.

The high bandwidth arbitrary transmit signal is generated with a multistage RF lineup from the DDS in-phase and quadrature (I/Q) components, through a quad DAC with quadrature modulator correction and group delay correction enabling IQ compensation for gain, offset, phase and group delay between channels, a quadrature up-converter to 5.5 GHz and then a multi-channel multiplier to 22 GHz for distribution to the Sivers BFM devices.

Coherence, phase noise and Allen variance are critical in a radar at this frequency, particularly when used in a SAR mode. The primary reference is an ultra-low jitter oscillator. This reference is provided to an ultra-low noise clock jitter cleaner with dual-loop phased-locked loops to distribute the multiple coherent clocks to various subsystems.

SAR systems are typically deployed on larger drones or aircraft. SAR has been demonstrated by researchers on small drones (< 50 kg), but space varying errors due to aerodynamic turbulence and flight path position and attitude errors degrade resolution and are an obstacle to deployment on sUAVs. The aiRadar InSAR has an innovative solution to this, using the multi-aperture SAR capability, DPCA micronavigation, dual GNSS (GPS) receivers, nine-axis attitude sensors and time domain back projection to support unsupervised SAR processing.

COMMAND AND TELEMETRY; THE AiRPL ECOSYSTEM

The SEM is controlled by the aiRadar aiRPL, a compiled language with a syntax like C, which runs on the aiRPU. This language provides sophisticated multilevel looping and calls to the transmit pulse modulations, the beamforming and the receive data processing. This system is a powerful yet simple tool for the programming of radar configurations and almost arbitrarily complex operational modes.

The aiRPL ecosystem is composed of an integrated software development environment with compiler, databases, a command processor and a radar precision timing processing unit. The integrated software development environment consists of two main parts:

  • The radar programming language—tools for the generation and sequencing of PRI bursts, sequences and frames.

Data structure creation and maintenance tools, including:

  • Transmit Pulse Design
  • Receive Configuration Design
  • Beamforming Design
  • TRM Hardware Configuration
  • Radar Constraint Definition
  • Image Quality Analysis
  • Test and Maintenance.


Built into the aiRPL are source methods for invoking radar operations and for sequencing these operations, for example, the PRI command. A PRI could be:

PRI (5e−3, “ tr0rx12,” “ bfwide7,” “ txfmcw4,” “rxfmcw1”)

In this example, a 5.0 msec PRI is programmed, accessing four structures, “tr0rx12,” “bfwide7,” “txfmcw4” and “rxfmcw1.” The first structure defines the Tx/Rx and interferometric configuration, the second controls the beamformer, the third defines the transmit pulse and the fourth defines the receive mode, digital filtering and decimation.

LPI RADAR

A fragment of radar programming code is shown below to demonstrate the programming of an LPI mode in which the radar transmitter has an LPI code hopping from PRI-to-PRI, with a triplet of pulses (two Frank codes and one Costas code) transmitted in each burst.

REPEAT(16) { // PRI sequence is executed 16x per burst

PRI(1.00E-03, “f_trcconf1,” “f_bfnoop0,” “f_txlpi0,” “f_rxconf0”) // Frank Code N = 3

PRI(1.00E-03, “f_trcconf1,” “f_bfnoop0,” “f_txlpi1,” “f_rxconf0”) // Costas Code array size = 10

PRI(1.00E-03, “f_trcconf1,” “f_bfnoop0,” “f_txlpi2,” “f_rxconf0”) // Frank Code with N = 4

}

A typical encoded LPI pulse in the frequency/time domain referred to in the code above as “f_txlpi0” is shown in Figure 6 (parameters have been chosen for graphics clarity) and Figure 7.

Figure 6

Figure 6 Typical LPI frequency vs. time code, spanning 25 MHz within a 20 µs period.

Figure 7

Figure 7 DDS output waveform with the typical LPI code.

IMAGE TEST RESULTS AND IMAGE QUALITY

aiRadar instruments provide ongoing image quality assessment tools to monitor performance by measuring quantitative performance parameters such as impulse response function, peak sidelobe ratio and ISLR.

Figure 8 is a screenshot from the aiRadar image quality analysis tool captured during preliminary calibration of the RRI-100 radar. It shows a point scatterer in a clutter rich short-range environment. Annotations on the image are added for clarity.

Figure 8

Figure 8 Screenshot from the image quality analysis tool.

Figure 9

Figure 9 Image of parking lots and rail yards.

Figure 9 shows an image of two parking lots with vehicles, dumpsters and a railway switching yard. A relatively shorter range of 50 m is selected to demonstrate filtering and down-sampling of the 20,000 range samples at 5 cm resolution. The radar was deployed at an elevation of 20 m with the elevation boresight horizontal.

MTI

ADAS requires excellent MTI processing to separate stationary infrastructure, such as buildings and traffic signs, from moving or stationary objects, such as cars, trucks, cyclists and pedestrians. V- and W-Band provide excellent sensitivity and resolution of moving objects. A stationary radar image with normal processing is shown in Figure 10a, the outlined area expanded in Figure 10b. The expanded area processed with 32 chirps in a frame and reprocessed with a 32-bin Doppler filter (i.e., 32-point FFT) clearly shows the moving target (see Figure 10c), which was invisible in the normally processed image.

Figure 10

Figure 10 Stationary radar image with normal processing (a), expanded view of highlighted area (b) and highlighted area with MTI processing (c).

CONCLUSION

aiRadar research radars facilitate the definition of validated requirements and AESA configurations for emerging commercial, military and academic radar applications. These research instruments provide the tools to validate requirements and develop sophisticated radar systems reducing time-to-market and offering a low risk path to commercialization and deployment. aiRadar offers in-house radar design for a customized application-specific radar or licensing of the aiRadar programming language (aiRPL) compiler and the radar processing unit IP Core (aiRPU). Developing compact low size, weight, power and cost, AESA radars with RAR, SAR, InSAR, multi-baseline MTI, LPI and CA has never been easier.