Beam Alignment and Tracking for Autonomous Vehicular Communication using IEEE 802.11ad-based Radar

Mobility scenarios involving short contact times pose a challenge for high bandwidth data transfer between autonomous vehicles and roadside base stations (BS). Millimeter wave bands are a viable solution as they offer enormous bandwidth in the 60GHz …

Authors: Guillem Reus Muns, Kumar Vijay Mishra, Carlos Bocanegra Guerra

Beam Alignment and Tracking for Autonomous Vehicular Communication using   IEEE 802.11ad-based Radar
Beam Alignment and T racking for Autonomous V ehicular Communication using IEEE 802.11ad-based Radar 1 Guillem Reus Muns, 2 Kumar V ijay Mishra, 3 Carlos Bocanegra Guerra, 4 Y onina C. Eldar and 5 Kaushik R. Cho wdhury 1 , 3 , 5 Electrical and Computer Engineering Department, Northeastern Univ ersity , Boston, MA, USA 2 , 4 Andrew and Erna V iterbi Faculty of Electrical Engineering, T echnion - Israel Institute of T echnology , Haifa, Israel Email: { 1 greusmuns@coe, 3 bocanegrac@coe, 5 krc@ece } .neu.edu, { 2 mishra, 4 yonina } @ee.technion.ac.il Abstract —Mobility scenarios in volving short contact times pose a challenge for high band width data transfer between autonomous vehicles and roadside base stations (BS). Millimeter wav e bands ar e a viable solution as they o ff er enormous bandwidth in the 60GHz band with several Gbps data transfer rates. Howe ver , beamforming is used as a default mode in this band, which requir es accurate and continuous alignment under relati ve motion. W e propose a method in which an o ff -the-shelf IEEE 802.11ad WiFi router is configured to serve as the BS as well as a radar exploiting special structure of 802.11ad preamble. W e embed the radar functionality within standards-compliant operations that do not modify the core structure of the frames beyond what is defined by the 802.11ad protocol. This not only reduces the beam training time, but also ensures scalability with increasing vehicular tra ffi c because radar allows accurate ranging of up to 0.1m at distances up to 200m. W e further analyze the ensuing cost-benefit trade-o ff between the time allotted to the proposed in-band radar and communication modes. Our r esults re veal 83% reduction on the overhead incurr ed during the beam training achieved for a specific simulated vehicular scenario ov er the classical 802.11ad operation. Index T erms —IEEE 802.11ad, mmW av e, automotive radar , beam alignment, CA V I. I ntr oduction Connected and autonomous v ehicles are being gradually integrated into consumer-le vel road transportation giv en the associated cost / energy sa vings [1], as well as increase safety and comfort. In addition, se veral companies use the same technology to generate digital maps and real-time 3-D visuals through real-time sensing, which can improv e o ffl ine path planning algorithms [2]. Each of these examples requires e ffi cient relaying of lar ge volumes of data by the vehicles, ranging in the order of several gigabits-per-second. The ke y problem that we address in this paper is ho w to devise a cost-e ff ectiv e communication architecture for vehicle-to-infrastructure (V2I) that o ff ers high data capacity while being resilient to the challenges posed by mobility . Millimeter W av e (mmW av e) technology in the recently opened 57 − 64 GHz band is a promising candidate for our scenario because of the av ailability of massi ve unlicensed spectrum. Consumer-e xploitation of this band is already underway: The IEEE 802.11ad standard enables throughput of up to 7 Gbps at 60 GHz for short-range wireless communication [3]. Although the available bandwidth is considerably wider in these bands (for e.g., IEEE 802.11ad defines up to 2 GHz wide channels), transmissions su ff er from Fig. 1. V2I scenario where a BS uses radar to localize and serve vehicles moving right to left. Radar operations (purple) allow for determining the location and speed, enabling directiv e and robust data transfer (green). These operations are scheduled within a standard MA C frame giv en in Fig. 2. high attenuation associated with increasing carrier frequency and channel losses arising from natural phenomena such as atmospheric absorption. In order to o vercome these problems, highly directional antennas are used. Here, the radiation beam patterns are configured to ensure the peak of the transmitted RF energy appears along the intended direction, while lowering the emissions in other locations. • Challenges in beamforming for mobile scenarios: First, narrow beams e ffi ciently focus energy in the chosen direction, but are less robust to movement, i.e. any beam misalignment causes a sharp loss in connecti vity . Second, the beam training procedure is time-consuming and lowers e ffi cienc y by utilizing network resources that could be used for communication. This worsens in mobile scenarios where the beam training function calls are frequent [4]. T raining in volves two major steps: through per-sector transmissions, users first locate each other in space during coarse-grained training, follo wed by exploration of the identified sector fine-grained training to lower the beamwidth. Finally , the use of narrow directional beams makes the connection sensitiv e to blockage, which is a significant problem in mmW av e compared with sub-6 GHz Fig. 2. The 802.11ad MA C frame configured to accommodate radar operations. R1 and R2 represent the radar periods while Tx1 and Tx2 are for the vehicle data transfer . The details of the physical frame (bottom) IEEE 802.11ad MAC header (top) are described in Sections III and V, respecti vely . bands. In most mobility scenarios, contact times are expected to be short. Hence, if standards-compliant protocols are used, classical beam training techniques are not feasible. • Proposed Appr oach: Our approach performs both radar operations and conv entional communication using a common transmitter / receiv er (Tx / Rx) chain (Fig. 1) with 802.11ad physical layer (PHY) frames, as proved earlier in [5, 6]. Howe ver , it extends further to the Media Access Control (MA C) layer by inte grating these functions within defined MA C protocol headers (see Fig. 2), but does not require any protocol modification. This approach quickly , and in a scalable manner , identifies the locations of moving vehicles, which in turn cuts down the time needed for classical beam training. W e address a number of technical issues to achieve this design: (i) incorporating the radar capability within the constraints and features of the protocol standard, (ii) analyzing and optimizing the durations allotted to radar and communications, and (iii) mapping the communications technology advancement to the operational demands of autonomous vehicles. Our main contributions are: 1) W e design and profile a 802.11ad PHY -based radar in terms of Doppler resolution, which determines the accuracy of the radar , its ability to track the vehicles, and the required time of operation. 2) W e connect an in-band radar ranging method with the classical 802.11ad operation, and study the associated cost-benefit trade-o ff s among the e ffi ciency of channel utilization, throughput, beam dimensions, among others. 3) W e propose a 802.11ad MA C configuration to accommodate radar operations alongside regular communications operations, without any change in the channel access operation. Consequently , this reduces misalignment in a high-mobility en vironment while maintaining low o verhead. 4) W e provide spatial characterization of the beamwidth, angular sector and sector overlap as a function of the radar performance, and analyze the resulting throughput when in-band radar and communication operations are performed using the same hardware. This paper is organized as follows. Section II revie ws the state of the art. Section III describes the technical details of the proposed 802.11ad-based radar , which influences the parameters used in the later sections. Section IV contains a detailed analysis of the achiev able throughput. In Section V, we provide an operational description of the vehicular scenario and connect the technical development to the application scenario. Section VI gi ves the performance e valuation and, finally , Section VII concludes the paper . II. R ela ted W ork Many recent research e ff orts ha ve been dev oted to reducing the beam training overhead in the mmW ave band. In [7], the authors reduce the time for beam training using in-packet BF . In [8], a codebook design scheme enhances the system performance. The 60 GHz and legac y 2.4 / 5 GHz bands are combined in [9] to accelerate the beam training. While the abov e mentioned works reduce the overhead of beam training, they are not designed for mobility situations. Recent work in volving mobility lev erages out-of-band position reports to aid in beam training [4, 10]. Similar de velopments hav e been published for high speed trains [11]. Howe ver , this work relies on position reports supplied through other bands to facilitate the directional communications. On the other hand, we use the same (IEEE 802.11ad) band to obtain such reports. The IEEE 802.11ad physical layer transmits control (CPHY), single carrier (SC) and orthogonal frequency-di vision multiplexing (OFDM) modulation frames at chip rates of 1.76 GHz and 2.64 GHz, respectively . Each IEEE 802.11ad CPHY and SCPHY frame consist of a short training field (STF), a channel estimation field (CEF), header , data and a beamforming training field (see Fig. 2). The CPHY and SCPHY frames have the same chip rate, carrier frequency and CEF structure. Ho we ver , the CPHY STF is longer than that of the SCPHY . The STF and CEF together form the corresponding (CPHY or SCPHY) preambles. Within the CEF lie two 512-point sequences G u 512 [ n ] and G v 512 [ n ] which encapsulate Golay pairs. These paired sequences hav e the property of perfect aperiodic autocorrelation (zero sidelobes) making them useful for communication channel estimation [5] and radar remote sensing [6]. In this paper , our radar employs 802.11ad CPHY / SCPHY frames to estimate the range and Doppler velocity of the v ehicles. Among prior studies, the closest to our work is [12] which employs radar for the beam-alignment but it does not specify any particular protocol for communication. In addition, it excludes high-mobility vehicles and the radar does not co-exist within the same band. In [6], an in-band radar that uses 802.11ad signaling is presented for v ehicle-to-vehicle (V2V) applications. The model assumes a single target and estimates its range using SCPHY STF and symbol boundary detection (SBD), while a coarse estimation of Doppler velocity is obtained using the pulse-pair method [13, 14]. Similar single-target analyses with CPHY STF are described in [15]. V ery accurate estimates of radar range and Doppler velocity are obtained in [16] using a 5 . 89 GHz dedicated short-range communication (DSRC)-based V2V communication link. Howe ver , the signaling is based on the 802.11p OFDM protocol where, for the radar operation, the subcarrier spacing should be more than the maximum Doppler shift, and the cyclic prefix duration needs to be greater than the longest round-trip delay . It is di ffi cult to maintain these requirements in a relati vely narro w 10 MHz DSRC band. In [17], a general-purpose spectrum sharing radar is presented using the recently proposed concept of cogniti ve radar that e xploits sub-Nyquist processing [18]. The automotiv e radar application modifications are suggested in [19] b ut these works employ IEEE 802.22 cognitiv e radio for communications. In contrast to these works, we use 802.11ad-based pulse-Doppler radar for beam alignment and radar network synchronization without any restriction on mobility . Unlike [6, 15], our model and processing are also applicable for both CPHY and SCPHY frames. III. 802.11 ad V2I A utomo tive rad ar Our range and Doppler estimation methods closely follow the CEF-based communication channel estimation described first in [5, 6] and, hence, we only summarize them here. A. Automotive radar signal pr ocessing A Golay complementary pair consists of two sequences G a N and G b N both of the same length N with entries ± 1. Consequently , their aperiodic autocorrelation functions hav e sidelobes equal in magnitude b ut opposite in sign. Therefore, the sum of the two autocorrelations has a peak of 2 N and a sidelobe lev el of zero: G a N [ n ] ∗ G a N [− n ] + G b N [ n ] ∗ G b N [− n ] = 2 N δ [ n ] , (1) where ∗ denotes linear conv olution. The 802.11ad CEF transmit signal is a concatenated sequence: s T [ n ] = Gu 512 [ n ] + G v 512 [ n − 512 ] , n = 0 , 1 , · · · , 1023 . (2) Here, the sequences G u 512 [ n ] and G v 512 [ n ] are defined for 0 ≤ n ≤ 511; for other values of n , they are set to zero. Each of them contains a Golay complementary pair of length 256, { G au 256 , G bu 256 } and { G a v 256 , G b v 256 } , respectiv ely . The discrete-time sequence s T [ n ] is passed through a digital-to-analog-con verter (D AC) the output of which can be represented as a weighted sum of Dirac impulses s T ( t ) = Í m s T [ m ] δ ( t − mT c ) . The 802.11ad protocol specifies a spectral mask for the transmit signal to limit inter-symbol interference (ISI) [3, section 21.3.2]. W e assume that h T ( t ) includes a lo w-pass baseband filter with an equiv alent amplitude characteristic of the spectral mask. A common shaping filter has a frequency response H T ( f ) of the root raised cosine (RRC) filter . The receiv er employs another RRC filter H R ( t ) such that the net response is equal to a raised cosine (RC) filter , H ( f ) = H T ( f ) H R ( f ) . The RC filter is a Nyquist filter with the following property to a void ISI: h [ n ] = h ( n T c ) =  1 , n = 0 0 , n , 0 . (3) W e can formulate this as: h ( t ) + ∞ Õ k = −∞ δ ( t − k T c ) = δ ( t ) . (4) This property only holds for the RC, and not the RRC filter . The baseband signal is then upconv erted for transmission: x ( t ) = x T ( t ) e j 2 π f c t , where f c denotes the carrier frequency . Suppose the radar transmits P packets at the pulse repetition interval (PRI) T p tow ards L direct-path nonfluctuating tar gets of complex reflectivity α l located at range d l = c τ l / 2 and Doppler velocity ν l = 2 π f D l , where c = 3 × 10 8 m / s is the speed of light, τ l is the time delay , and f D l is the associated Doppler frequency . Ignoring the multi-path components (MPC), the reflected receiv ed signal at the baseband, i.e., after down-con version, is giv en by x R ( t ) = P − 1 Õ p = 0 L Õ l = 1 α l x T ( t − τ l − pT p ) e − j 2 π f D l t + z ( t ) ≈ P − 1 Õ p = 0 L Õ l = 1 α l x T ( t − τ l − pT p ) e − j 2 π f D l pT p + z ( t ) , (5) where z ( t ) is additive circular-symmetric white Gaussian noise. The last approximation follows from the fact that f D l  1 / T p so that the phase rotation within one coherent processing interv al (CPI) (“slow time”) can be approximated as a constant. Here, we have assumed that the coe ffi cient α l also characterizes antenna directivity , processing gains and attenuations including path loss. The receiv ed signal is sampled at F c = 1 / T c to obtain x R [ n ] = x R ( nT c ) = P − 1 Õ p = 0 L Õ l = 1 α l x T ( nT c − τ l − pT p ) e − j 2 π f D l pT p + z ( nT c ) = P − 1 Õ p = 0 L Õ l = 1 α l s T ( nT c − τ l − pT p ) e − j 2 π f D l pT p + z [ n ] , (6) where we used the Nyquist filter properties (3)-(4) to obtain the last equality . The sampled signal is passed through matched filters of each Golay sequence e.g. for the first pair , we have ˆ h 1 au [ n ] = x R [ n ] ∗ G au 256 [− n ] ˆ h 1 b u [ n ] = x R [ n ] ∗ G bu 256 [− n ] . (7) These outputs are delayed: ˆ h 1 a [ n ] = ˆ h 1 au [ n ] , ˆ h 1 b [ n ] = ˆ h 1 b u [ n + 256 ] , and summed up to yield the channel estimate ˆ h 1 [ n ] = 1 512 ( ˆ h 1 a [ n ] + ˆ h 1 b [ n ]) ≈ 1 512 P − 1 Õ p = 0 L Õ l = 1 α l δ ( n T c − τ l − pT p ) e − j 2 π f D l pT p + z [ n ] ∗ ( G a u 256 [− n ] + G bu 256 [− n ]) , (8) where we assumed that the Doppler shifts are nearly identical for the two Golay sequences G au 256 and G bu 256 enabling us to use the Golay pair property (1). Another estimate ˆ h 2 [ n ] of the radar channel is available via the second Golay pair within G v 512 . An average of the two estimates yields a final approximation of the radar channel. W e discretize the range-time space in, say , N R bins of resolution cT c / 2. W e then create a delay-Doppler map by taking a P -point Discrete F ourier T ransform (DFT) of the radar 1 1.5 2 2.5 3 PRI Length (s) 10 -4 15 20 25 30 35 40 45 50 Max Velocity Detection (m/s) Max. Velocity vs PRI Length Max Vel Expected Max Vel (a) 20 40 60 80 100 Number of Packets (P) 0.5 1 1.5 2 2.5 Doppler resolution (m/s) Radar velocity accuracy (m/s) PRI = 100us PRI = 120us PRI = 140us PRI = 166us (b) Fig. 3. (a) Maximum detectable velocity ν u as a function of the PRI length. For highways, where ν u = 30 m / s, PRI shall be no greater than 0.166 ms. (b) The radar accuracy as a function of P for PRI no greater than the highway limit. channel estimates corresponding to each delay bin. Then, the delay and Doppler frequencies of the tar gets are gi ven by the location of the first L peaks on this 2D delay-Doppler map using e.g. a constant false alarm rate (CF AR) detector . In a pulse Doppler radar , the abo ve-mentioned processing estimates only the radial component of the vehicle’ s v elocity . If needed, a radar can estimate the tangential component by measuring target’ s motion on a clutter map. B. V2I radar design At the carrier frequency f c of 60 GHz, the wa velength is λ c = c / f c = 0 . 005 m. The STF of CPHY and SCPHY frames contain 6400 and 2176 symbols, respectiv ely . The CEF of both frames has 1152 symbols each. For the symbol time of T c = 1 / 1 . 76 e 9 = 0 . 568 ns, we obtain a minimum PRI of T pr = T c ∗ ( 6400 + 1152 ) = 4 . 29 µ s for CPHY while the same is 1 . 89 µ s for the SCPHY . A typical long range V2I radar (LRR) operates with a maximum unambiguous range of R u = 200 m and range resolution of 0 . 1 m. The symbol time of 802.11ad yields a good range resolution of ∆ R = cT c / 2 = 8 . 52 cm. If P packets are transmitted, then the CPI duration is T int = PT pr . For pulse repetition at a uniform rate, the maximum unambiguous Doppler velocity of the radar is giv en by ν u = λ c / T pr and the Doppler resolution obtained through Fourier processing is ∆ ν = λ c /( 2 T int ) . Fig. 3 sho ws the dependence of the Doppler resolution on the number of packets and PRI. The total time the radar requires to scan a certain radar sector can be expressed as: t radar ≈ T int b ϕ Sradar / θ radar c , (9) where b · c denotes the floor function, θ radar is the scan rate per unit CPI, fixed to 0 . 5 ◦ / CPI, and ϕ Sradar is the angular sector size. IV . 802.11 ad PHY / MA C P ro tocol C onfigura tion Consider a BS located at distance d from an ideal straight road, serving V number of non-stationary vehicles that pass along a fixed and uniform route (Fig. 4). Each vehicle c l is characterized by its true location and velocity at time t , y l ( t ) and ν l ( t ) , respectiv ely . The BS obtains the radar estimates ˆ d l ( t ) and ˆ ν l ( t ) ) of the location and velocity , respecti vely . For simplicity , ˆ ν l ( t ) will be assumed to be constant. As we describe Fig. 4. Example of beam design with four beam. below , the 802.11ad throughput is a ff ected by the location of the vehicles with respect to the BS. A. Link Budget Model Giv en the communication scenario (Fig. 1) and transmit power P tx , we assume line-of-sight (LOS), allo wing for the following simplified expression for the receiv ed power P rx [4] (all quantities in dB): P rx = P tx + G tx − P L + G rx , (10) where P L is the path loss and G tx and G rx are the antenna gains in transmission and reception, respecti vely . Here, the path loss is giv en by P L = 10 n log 10 d v + S F + C att + A att + R att , (11) where n is the path loss exponent, d v is the distance from the Tx in the BS to the vehicle, and S F corresponds to the random shadowing e ff ect following a lognormal distribution S F ∼ N ( 0 , σ 2 ) with σ = 5 . 8. The C att , A att and R att are the LOS link, atmospheric and rain attenuations, respectiv ely . The linear scale antenna gains are given by G tx / rx = 4 · 180 2 θ el θ az π , (12) where θ el and θ az are the half-power beamwidths (radians) in elev ation and azimuth for the corresponding antennas. T o be consistent with the previous work in [4] and [10], we assumed an ideal beamforming in (12), characterized by uniformly distributed gain within the beam and no side lobes (ideal antenna e ffi ciency). W e model the impairments at the receiv er by accounting for the phase noise P noise as follows: P noise = N floor + 10 log 10 B + NF , (13) where N floor is the noise floor, B is the system bandwidth and N F the noise figure of the receiv er chain. As a result, the SNR is a function of the Tx-Rx distance d v and beamwidth: SNR ( θ az , d v ) = P rx ( θ az , d v ) P noise . (14) Here, θ el is fixed to a value that encompasses the width of the road. Hence, it depends on the distance d . Both θ el and θ az are assumed identical in transmission and reception. B. A verage Rate For each Modulation and Coding Scheme (MCS), the 802.11ad standard requires a minimum Packet Error Rate (PER) which, in turn, is inherently impacted by the Signal-to-Noise Ratio (SNR). As in (14), the SNR itself depends on the scenario geometry d v and θ az . W e can express 20 40 60 80 100 4000 4100 4200 4300 4400 4500 4600 4700 Throughput vs distance Beamwidth = 2° Beamwidth = 3° Beamwidth = 5° Beamwidth = 10° (a) 10 20 30 40 1500 2000 2500 3000 3500 4000 4500 Throughput vs angle Distance = 10 (m) Distance = 50 (m) Distance = 75 (m) Distance = 100 (m) (b) Fig. 5. A verage throughput analysis with respect to the (a) BS-Road distance and (b) beamwidth. this dependenc y as a function SNR ( θ az , d v ) . Further , the MCS-0 (most robust) demands a PER lower than a threshold ( γ ) of 5% for a PHY service data unit (PSDU) length of 256 octets. For any other MCS, the PER shall be less than γ = 1% for a PSDU length of 4096 octets [3]. Thus, the achiev able throughput R 802 . 11 a d is characterized by the tuple { SNR,PER,MCS } . T o obtain the maximum data rate R , our approach always selects the MCS corresponding to the highest data rate among the ones that meet the PER requirements: R ( θ az , d v ) = max MCS PER ( SNR ( θ az , d v )) ≤ γ R 802 . 11 a d ( MCS ) . (15) The av erage data rate R l , or simply R , between, say , a l th vehicle and the BS is computed as the integral of the achiev able rate over the contact time interval [ t 1 , t 2 ]: R = 1 t 2 − t 1 ∫ t 2 t 1 R ( t ) d t . (16) For ease of notation, we rewrite [ t 1 , t 2 ] as [ t init , t init + t c ]: R = 1 t c ∫ t init + t c t init R ( t ) d t , (17) where t init corresponds to the instant when the l th v ehicle enters the area cov ered by the BS and t c is the contact duration: t c = 2 d tan  φ BS 2  ν l , (18) where φ BS is the BS cov erage angle (see Fig. 4). From Fig. 5, where R is plotted as a function of d and θ az , we conclude following: First, a narrower beamwidth helps mitigate the high path loss attenuation, achieve a higher SNR and, in turn, increase the achiev able throughput. Second, the achiev able throughput is impacted more by the beamwidth θ az than the distance d . The latter finding serves as a design constraint in our system, i.e., narr ow beamwidth is pr eferred over short distances . V . 802.11 ad - based J oint rad ar and C ommunica tion In this section, we discuss the intervals of calling the radar function, and the time alloted to complete this function per call. Beginning with the fields used in the 802.11ad standard, we identify the elements in the header that should be configured to incorporate the radar operation. A. Beam alignment model Consider the scenario in Fig. 4. The spatial orientation of the beams follow the beam-switching pattern [11]. Only one sector is activ ated at a time, and the BS either performs the radar or communication operations [10]. Ho wever , the respectiv e sectors are defined di ff erently: For communication, the adjacent sectors hav e some ov erlap that is specified by an ov erlap ratio parameter . The radar sectors do not ov erlap and their angular aperture is controlled by ϕ S r a d a r . As stated in Section IV -A, the beams are assumed to be ideal, i.e., if at any instant a vehicle is not within the limits of the active sector , we consider the beams to be in complete misalignment. B. Using 802.11ad Headers The 802.11ad protocol structures its frames into Beacon Intervals (BI) (Fig. 2) with 30 ms duration for vehicular networks [4]. The BIs consist of Beacon Header Interval (BHI), where antenna beams are formed and network synchronization information is exchanged, and Data Transfer Interval (DTI), during which the actual communication occurs. On top of this, the protocol also defines a Beam Refinement Protocol (BRP) during the DTI, where a similar procedure is carried out with narrower beams. Three subintervals constitute BHI in the following order: • Beacon T ransmission Interval (BTI) . The BS uses beacon frames to announce its presence, scan the whole cov erage area using S I sectors, allow the de vices to align their beams and acquire basic network information. This is the initiation phase in a Sector Lev el Sweep (SLS). • Associated Beamforming T raining (A-BFT) . The de vices contend in a random slotted access fashion upon completion of the BTI and the Medium Beamforming Interframe Spacing (MBIFS). The devices first use S R sectors to reply to the BS, make way to a later feedback message from the BS and send the ackno wledgement (A CK) that closes the SLS (Response Phase). • Announcement T ransmission Interval (ATI) . T o allocate the devices during the DTI, the BS updates the number of connections and inquires them about their data needs. W e explain our PHY -MA C configuration through a scenario requiring vehicles to synchronize with a gi ven BS that has radar capabilities. First, the BS sends P Directional Multi-Gigabit (DMG) Beacon frames (see Section III-B) through S I = ϕ Sradar / θ radar sectors within a confined area (shown by Radar in Fig. 1), allowing the BS to obtain an estimation of the locations and velocities while providing the vehicles with synchronization information. The duration of the radar slot is characterized in (9) and (23). At the end of the BTI and after the radar call, the BS detects each car location, and then sends a DMG beacon to each vehicle assigning them to an A-BFT slot (in order to avoid ine ffi cient contention during legacy 802.11ad). The BS antenna is also included in this message, which allo ws the vehicles to calculate the best transmit sector . Upon completion of the BTI, vehicles start the Response Phase and transmit through their S R sectors. Howe ver , they will already kno w the BS location, therefore S R is equal to 1. In summary , we perform two steps with only one transmission (radar and synchronization) during the BTI. T ABLE I P arameter v alues for numerical experiments Parameters V alue T ransmission power P t x 10 dBm Path-Loss exponent n 2.66 Channel attenuation C att 70 dB Atmospheric attenuation A att 15 dB km − 1 Rain attenuation R att 25 dB km − 1 Noise Floor N floor -174 dBm Bandwidth B 2.16 GHz Carrier frequency f c 60 GHz W avelength λ c 5 mm Noise figure N F 6 dB Max. V elocity v max 30 m / s BS coverage angle φ BS 120 degrees BS-Road distance d 100 m Communications beamwidth θ az 3 degrees Outage 2 % Initiator number of sectors (802.11ad) S I 32 Responder number of sectors (802.11ad) S R 32 Number of A-BFT slots 4 Furthermore, we reduce the A-BFT slots by confining the BTI within the area defined by ϕ S r a d a r and eliminate the BRP through the high localization accuracy acquired by the radar in the DTI period. C. Radar comprehensive configuration - system design The radar detects vehicles passing by a pre-defined sector and estimates their position and v elocity , ˆ d l ( t ) and ˆ ν l ( t ) , at a given time t . In this section, we detail (i) setting up the periodic sweep interval between consecuti ve sweeps, (ii) selecting sectors for the radar sweep, and (iii) allotting the time per sweep, i.e., decide the number of packets P that should be transmitted to achiev e a desirable accuracy . • Sweep interval: The minimum elapsed time, T radar , i , between two consecutiv e sweeps across the same i th sector , must be set such that ev ery vehicle which passes over that sector is detected. T radar , i (19) is restricted by the sector length, r i , and the maximum velocity expected in a vehicle, v max . T radar , i ≤ r i v max . (19) • Sector location: The radar sector length, r i (20), is defined as the road length co vered by the radar , which can be e xpressed as a function of ϕ Sradar , φ BS and the radar sector starting angle, θ start (see Fig. 1). r i = d  tan  1 2 φ BS − θ start  − tan  1 2 φ BS − ( θ start + ϕ Sradar )   . (20) W e simplify the analysis by restricting the number of radar sectors to one. Thus, θ start and ϕ Sradar hav e to be determined for that single sector . Also, as vehicles must be detected as soon as they enter the coverage area for e ffi cient resource utilization, consider the case where the radar sector is defined by θ start = 0. Now , the system needs to impose ne w restrictions tow ards setting ϕ Sradar : (i) the sector length covered by radar is not smaller than the length of the vehicle, w car (21), and (ii) the distance that a vehicle may mov e during a radar sweep cannot be larger than a set value ( k 2 ) (22). The location of the l th vehicle at a giv en time t is denoted by y l ( t ) . T wo qualitative constraints defined by constants k 1 and k 2 are included below for flexibility in the analysis. 5 10 15 20 25 30 35 40 Radar Sector (degrees) 4 6 8 10 12 14 16 Radar percentage [%] Overhead analysis vs radar sector start = 0º start = 15º start = 30º start = 50º start = 65º start = 80º Fig. 6. Radar percentage for di ff erent starting angles. ∆ ν = 0.456 m / s is used. r > k 1 · w car , (21) [ y l ( t + t radar ) − y l ( t )] < k 2 . (22) Consider now that the radar is configured to scan the whole cov erage area at once. If we want to achiev e a Doppler resolution of 1.5 m / s, incurring on a 5% error on the velocity estimation in the worst case, the system would require around 400 ms to complete the beam alignment procedure. This is not viable, since a vehicle that was detected at the very beginning would hav e covered 12 m upon completion of the radar function (22). Therefore, the radar location design took a less simplistic approach, emplo ying the radar only in a reduced area. Consider a ϕ Sradar of 5 ◦ , a ∆ ν = 0 . 454 m / s and a w car ≈ 5m. This values would correspond to r ≈ 30 m ( ≈ 6 w car ) and an in-radar-measure moved distance around 1.75 m ( ≈ w car / 3). This is a numerical example of fair values for ϕ Sradar and ∆ ν = 0 . 454 m / s according to the qualitati ve constraints (21) and (22). • Sweep time: W e define ρ as the ratio between the time taken to perform a full sweep of the chosen sector, t radar , and the ov erall interval between sweeps T radar : ρ = t radar T radar , (23) ρ ( ϕ Sradar , θ start ) = ϕ Sradar · λ c · v max ∆ ν · r ( ϕ Sradar , θ start ) , (24) where the last equality is obtained by substituting (9) and (19) into (23). VI. P erformance E v alua tion Consider the scenario defined in Section V -A and represented in Fig. 4, where vehicles will synchronize with the BS through the procedure defined in Section V -B. W e simulated this operation in MA TLAB using the PHY parameters defined in the WLAN Systems T oolbox and T able I. W e compared the overhead between the legac y 802.11ad and our modified configuration using 802.11ad-based Radar . Fig. 6 shows ρ as a function of ϕ Sradar for di ff erent fixed initial locations θ start . The configuration that provides highest 40 60 80 100 120 Angle (degrees) 0 0.2 0.4 0.6 0.8 1 CDF Misalignment with Overlap ratio = 0.7 = 0.45 (m/s) = 0.5 (m/s) = 0.55 (m/s) = 0.6 (m/s) = 0.75 (m/s) = 1 (m/s) (a) 40 60 80 100 120 Angle (degrees) 0.2 0.4 0.6 0.8 1 CDF Misalignment with = 0.55 m/s Overlap ratio = 0.8 Overlap ratio = 0.7 Overlap ratio = 0.6 Overlap ratio = 0.5 Overlap ratio = 0.4 Overlap ratio = 0.3 (b) Fig. 7. Misalignment cumulative probability density function (CDF) for (a) di ff erent Doppler resolutions and (b) ov erlap ratios. The outage tolerance is 2%. 2 4 6 8 10 12 Radar Sector (degrees) 0 50 100 150 Time for Beamforming (ms) Beamforming time = 0.45 (m/s) = 0.5 (m/s) = 0.55 (m/s) = 0.6 (m/s) = 0.75 (m/s) = 1 (m/s) 802.11ad (a) 0 20 40 60 80 100 Radar Sector (degrees) 0 10 20 30 40 [%] Overhead analysis vs radar sector = 0.45 (m/s) = 0.5 (m/s) = 0.55 (m/s) = 0.6 (m/s) = 0.75 (m/s) = 1 (m/s) 802.11ad (b) Fig. 8. (a) The time incurred for beamforming in one BHI worsens with radar . (b) The time spent performing beamforming is reduced because of its longer periodicity . radar e ffi ciency by minimizing ρ is θ start = 0. As mentioned in Section V -C, this aids in the detection of cars as soon as they enter the coverage area. Therefore, placing the radar sector at the edge of the cov erage area is the optimal choice. As stated in Section V, we employ beam-ov erlapping to mitigate potential radar estimation errors. In Fig. 7, the impact of the Doppler resolution ( ∆ ν ) and beam-ov erlap on the misalignment probability is presented. Accurate velocity estimations require lo w ∆ ν . Howe ver , as seen in Fig. 3, improv ed radar accuracy implies longer radar transmission periods. W e now study the beamforming time to observe the reduction in the ov erhead o ver baseline 802.11ad. Fig. 8a shows the time a single beamforming procedure requires in terms of ϕ Sradar . Even if the radar increases the beamforming time for a single BHI, as per (19), we only repeat this procedure every T radar . In contrast, legac y 802.11ad for V2I requires such a repetition every ≈ 30 ms [4], which implies ≈ 35% of the overall access time. Fig. 8b shows the percentage time (24) our configuration requires of the total time. Consider a radar design with a ∆ ν of 0 . 45 m / s and a ϕ Sradar of 5 ◦ . For a cumulative misalignment probability of less than 1% at the end of the coverage area, an overlap ratio of 0.7 is employed. This implies a beam training ov erhead in the BHI of 59 ms in contrast to the 10 . 72 ms of the 802.11ad. Ho wever , our design introduces a total overhead of 5.82% compared to the 35% from the standard. Thus, we achieve a reduction of 83% on the 802.11ad beam training ov erhead. VII. C onclusion This work demonstrates the feasibility for using radar and communications jointly within a single 802.11ad transcei ver chain for high bandwidth and mobility situations. W e configured the 802.11ad MA C to embed radar operations with standards-compliant packets and at the same time, improve upon the completion time and scalability of legac y beam training procedures. W e have studied the ideal parameter settings for the radar operation, including the impact of v arying the sweep ov erlap and the achiev able Doppler resolution. A cknowledgments Kumar V ijay Mishra acknowledges partial support via Andrew and Erna Finci V iterbi Fellowship and Lady Davis Fellowship. R eferences [1] K. N. Qureshi and A. H. Abdullah, “ A survey on intelligent transportation systems, ” Middle-East Journal of Scientific Researc h , vol. 15, no. 5, pp. 629–642, 2013. [2] J. Le vinson, J. Askeland, J. Becker , J. Dolson, D. Held, S. Kammel, J. Z. K olter, D. Langer , O. Pink, V . Pratt et al. , “T owards fully autonomous dri ving: Systems and algorithms, ” in IEEE Intelligent V ehicles Symposium , 2011, pp. 163–168. [3] “Wireless LAN Medium Access Control (MA C) and Physical Layer (PHY) Specifications Amendment 3: Enhancements for V ery High Throughput in the 60 GHz Band, ” IEEE Std. 802.11ad , 2012. [4] I. Mavromatis, A. T assi, R. J. Piechocki, and A. Nix, “Beam alignment for millimetre wav e links with motion prediction of autonomous vehicles, ” arXiv preprint , 2017. [5] K. V . Mishra and Y . C. Eldar, “Sub-Nyquist channel estimation over IEEE 802.11ad link, ” in IEEE International Conference on Sampling Theory and Applications , 2017, pp. 355–359. [6] P . Kumari, N. Gonz ´ alez-Prelcic, and R. W . Heath Jr ., “Inv estigating the IEEE 802.11ad standard for millimeter wave automotiv e radar, ” in IEEE V ehicular T echnology Confer ence - Spring , 2015, pp. 1–5. [7] Y . M. Tsang, A. S. Poon, and S. Addepalli, “Coding the beams: Improving beamforming training in mmW av e communication system, ” in IEEE Global T elecommunications Conference , 2011, pp. 1–6. [8] Z. W eixia, G. Chao, D. Guanglong, W . Zhenyu, and G. Y ing, “ A new codebook design scheme for fast beam searching in millimeter-w av e communications, ” China Communications , vol. 11, no. 6, pp. 12–22, 2014. [9] T . Nitsche, A. B. Flores, E. W . Knightly , and J. Widmer , “Steering with eyes closed: Mm-W ave beam steering without in-band measurement, ” in IEEE Conference on Computer Communications , 2015, pp. 2416–2424. [10] V . V a, T . Shimizu, G. Bansal, and R. W . Heath Jr ., “Beam design for beam switching based millimeter wave vehicle-to-infrastructure communications, ” in IEEE International Conference on Communications , 2016, pp. 1–6. [11] V . V a, X. Zhang, and R. W . Heath Jr ., “Beam switching for millimeter wav e communication to support high speed trains, ” in IEEE V ehicular T echnology Conference - F all , 2015, pp. 1–5. [12] N. Gonz ´ alez-Prelcic, R. M ´ endez-Rial, and R. W . Heath Jr ., “Radar aided beam alignment in MmW ave V2I communications supporting antenna div ersity , ” in IEEE Information Theory and Applications W orkshop , 2016. [13] J. George, K. V . Mishra, C. M. Nguyen, and V . Chandrasekar , “Implementation of blind zone and range-velocity ambiguity mitigation for solid-state weather radar, ” in IEEE Radar Confer ence , 2010, pp. 1434–1438. [14] K. V . Mishra, V . Chandrasekar , C. Nguyen, and M. V ega, “The signal processor system for the NASA Dual-frequency Dual-polarized Doppler Radar , ” in IEEE International Geoscience and Remote Sensing Symposium , 2012, pp. 4774–4777. [15] E. Grossi, M. Lops, L. V enturino, and A. Zappone, “Opportunistic automotiv e radar using the IEEE 802.11ad standard, ” in IEEE Radar Confer ence , 2017, pp. 1196–1200. [16] D. H. Nguyen and R. W . Heath Jr ., “Delay and Doppler processing for multi-target detection with IEEE 802.11 OFDM signaling, ” in IEEE International Conference on Acoustics, Speech, and Signal Pr ocessing , 2017, pp. 3414–3418. [17] D. Cohen, K. V . Mishra, and Y . C. Eldar , “Spectrum sharing radar: Coexistence via Xampling, ” arXiv preprint , 2016. [18] O. Bar-Ilan and Y . C. Eldar, “Sub-Nyquist radar via Doppler focusing, ” IEEE T ransactions on Signal Processing , vol. 62, no. 7, pp. 1796–1811. [19] K. V . Mishra, A. Zhitnikov , and Y . C. Eldar , “Spectrum sharing solution for automotive radar, ” in IEEE V ehicular T echnology Confer ence - Spring , 2017, pp. 1–5.

Original Paper

Loading high-quality paper...

Comments & Academic Discussion

Loading comments...

Leave a Comment