Aerial distributed beamforming < 2 > Beamforming concept - - PowerPoint PPT Presentation

AirBeam: Experimental

Demonstration of Distributed Beamforming by a Swarm of UAVs

Next GEneration NEtworks and SYStems Lab

Presenter: Kaushik R. Chowdhury Authors: Subhramoy Mohanti Kumar Vijay, Carlos Bocanegra, Jason Meyer Gokhan Secinti† , Mithun Diddi, Hanumant Singh and Kaushik Chowdhury

> <

2

Aerial distributed beamforming

> <

3

Beamforming concept

The transmitters cooperatively organize themselves into a virtual antenna array and focus their

transmission in the direction of the receiver, such that they add up coherently at the receiver The transmitters make sure their transmission has a constructive interference effect at the sensor, which leads to a factor of N^2 gain in power efficiency, where N is the number of collaborating transmitters

Tx Tx Rx

> <

4

Traditional aerial network performance

2 hop topology with UAV as mid-hop router BER at receiver (B) with UAV (C) static (on ground) and hovering

> <

5

Static on-ground multi-user transmit beamforming testbed

> <

6

UAS test-bed setup to evaluate multi-user transmit beamforming

> <

7

Gold code sequence

Gold sequence, is a type of binary sequence, having bounded small cross-correlations within a set Gold codes have bounded small cross-correlations within a set, which is useful when multiple devices are transmitting in the same frequency range A set of Gold code sequences consists of 2n + 1 sequences each one with a period of 2n − 1

Tx Tx Rx Gold sequence 1 Gold sequence 2

> <

8

Probability of BER below X

Benefit of beamforming for disconnected aerial networks

> <

9

CSI estimation, BF weight calculation

The beamforming system can be modeled as:

y[m] = h^x[m] + n[m]

where h=[h1,...,hN] are the fixed channel gains from the transmit antenna to the receive antenna, and n[m] represents the additive white Gaussian noise (AWGN) at the receiver with a Normal distribution Aiming to achieve constant data-rate over time and targeting low latency applications, the system employs closed-loop Channel State Information (CSI) feedback where the receiver updates the transmitter constantly This way, the transmitted symbols s[m] are multiplied by the beamforming weights w as: is the average energy of the transmitted signal x[m] with normalized constellation symbols at any instant 1, with E being the mean function The weights are selected as to maximize the average mutual information function:

• ffering a total capacity:

> <

10

Beamforming with distributed aerial transmitters

> <

11

Challenges and Solutions

Sync problem weight restrictions for carrying heavyweight processors and bulky X310 radios Solutions: NVIDIA Jetson TX2 Ettus B210 SDR Fast processing in Gnuradio

> <

13

Relation between UAV hovering and channel gain

Gaussian distribution of UAV lateral displacement Channel gain with 1, 2 and 4 Tx UAVs

In the presence of classical GPS, the lateral displacement of the UAV is typically +/- 0.5m, even when the UAV is assigned to a fixed location in space Low channel gain fluctuations,can be observed during moments of high stability of the UAVs

> <

14

GnuRadio based aerial beamforming algorithm

Start Wait 50ms Payload Data BPSK QPSK 8-QAM 16-QAM 32-QAM 64-QAM Symbol generation Multiplex FFT Beamformed payload Multiplex USRP transmit Stop CSI feedback adapter Receiver feedback Zero padding Training signal

> <

15

UAV solution stack

> <

16

Faster receiver and wireless feedback

CSI estimation time in MATLAB, GnuRadio and fast processing enabled GnuRadio

> <

17

Pulse Per Second (PPS)

Some radio clocks and related timekeeping gear have a pulse-per- second (PPS) signal that can be used to discipline the local clock oscillator to a high degree of precision, typically to the order less than 50 us in time. PPS signal does not specify the time, but merely the start of a second. The pulse width is generally 100 ms, but many receivers allow the user to specify the pulse width, as long as it is less than one second.

> <

18

10 MHz reference signal

The 10 MHz frequency reference is used to phase lock the local oscillator, making the local oscillator frequency very stable. It ensures, the carrier burst arrives at the receiver at exactly the same frequency as the burst from another transmitter.

> <

19

Tx1 Tx2 Tx1 Tx2 PPS

Host system time synchronized with common clock reference (local server) PPS misalignment PPS aligned at the synced system clock reference point

Tx1 Tx2

Transmission starts after a global #seconds, which now coincides with the PPS start time

PPS misalignment of distributed B210s and solution

> <

20

Inaccurate cross-correlation due to PPS misalignment

Transmitters: 2 B210 radios each interfacing with their individual

Jetson TX2 host

Receiver: 1 B210 radio connected to another Jetson TX2 host Issue: Without any common time reference, and even with external 10

MHz reference and PPS from Octoclock, the distributed B210 radios were not able to beamform in a time coherent manner

Solution (Phase 1):

• One host elected to provide common time reference

to all distributed radios

• PPS of each radio hardware then locked on to this

common time reference

• Correlation inaccuracy brought down to 20us from

2000us

• But still not good enough for beamforming

> <

21

Degraded BER due to inaccurate cross-correlation

> <

22

Resolving cross-correlation issue

> <

23

BER target achievement

Probability of BER below certain value (x-axis) with various modulation schemes, 4 UAV-Tx and 1 ground-Rx.

> <

24

Conclusion

• With coherence gap of 1us between the four transmitted signals, faster GnuRadio based signal

processing and stable hovering of UAVs, cross-correlation was able to get the required starting index of the received payloads from the two signals

• Result was low BER in all the modulation schemes used, as given in the figure
• Beamforming was achieved with distributed low capability COTS radios
• Next Steps:
• Replacing the Octoclock with the GPSDO as the last step in truly wireless beamforming with distributed

radios

> <

Why do wireless radios have Clock Frequency Offset?

1PPS Reference Ideal 10 MHz/0pp b Faster +1ppb Slower

• 1ppb
• 100ns

+100ns

• 200ns

+200ns

26

> <

RFClock Receiver Two single-toned Transmission

Distributed RFClock Recipents

Distributed Clock Synchronization

Radio Radio Radio Radio

RFClock TX

Reference Clock

Amplifier

TX2 TX1

USRP Radio RFClock Transceiver

Clock Generation

Envelope Detection Removing DC Offset Converting Logic Level

Control Unit

Jitter Reduction

PPS Generation

TOF Measurement Delay Generation

RFClock RX USRP Radio

Reference and System Clock Generation

RF Front End

10 MHz 1 PPS

FPGA

27

> <

27

REAL TEST RESULTS: RFCLOCK

Clock Transmitter: Transmitting two-tone over air

Time Domain Behavior Clock TX:

• B210s were used to send two-tone signals at 900 MHz and 940

MHz with 40 MHz separation over the air. This separation freq can be chosen anything. (e.g 10 MHz) Stage 1

• Detect envelope of two-single tone signal by its separation

frequency as 40 MHz.

• Envelope detector sensitivity is -25 dBm, allows to large range

signal detection.

• Envelope signal peak-to-peak voltage level is about 1 V and

need to remove DC offset from the signal.

f1 – f2

Stage 1: Envelope Detection

> <

28

RFCLOCK TEST RESULTS

Stage 2: Removing Offset and Filtering

Stage 2

• Removing DC Offset from the envelope.
• Harmonics were filtered from the signal and distortion on the signal

were eliminated. Stage 3

• Translating sinewave signal to logic level with minimal additive

noise. Stage 4

• Frequency divider designed with flip flops and counters is used to

divide clock signal to generate 1 PPS signal.

Stage 4: PPs Generation Stage 3: Converting Logic Levels

> <

Control Logic Control Unit

MCU

Jitter Reduction 10 MHz SPI SPI

GPS PPS

PPS Generation

1 PPS Output

• Significant jitter reduction obtained at output of clock signal by using DSPLL with narrow loop filter bandwidth.
• Even if independent clocks have same clock rate (freq synchronization), each might have different start time, resulting

misalignment at the edges.

• Misalignment issue solved by delaying clock edge to common PPS signal.

RFClock Recipient

10 MHz Output 29 Clock Generation

Envelope Detection Removing DC Offset Converting Logic Level PD LPF %M

40 MHz

Frequency Divider

Jitter Reduction

Start Stop

Time-To-Digital Converter

Time of Flight Measurement Delay Line

Programmable Delay 8-bit latch

10 MHz

> <

30

Measurement of Clock Frequency Offset

TX RX RFClock RFClock

10 Mhz 1 PPS 10 Mhz 1 PPS

• Two USRP B210 nodes equipped with

RFClock recipients, one acting as a transmitter and the other as a receiver.

• The transmitter transmits packets

consisting of a single repeated OFDM symbol.

• The receiver computes its CFO with

respect to the transmitter using the traditional correlation-based OFDM CFO estimation Algorithm.

> <

31

Measurement of Clock Frequency Offset

> <

Distributed RFClock Setup

TX1 TX2 Amplifie r Reference UAV equipped with RFClock/B210 RFClock Transciever RFClock-1 RFClock-2

32

> <

Distributed Clock Synchronization with Beamforming

33