1 of 29

Quantum-Inspired Computing

for Large-Scale NOMA-MIMO Wireless Networks

Interns: Jeffrey Tang, Alex Markley

Advisors: Minsung Kim, Byungjun Kim

1

2 of 29

Project Team

2

Jeffrey Tang

Rutgers Undergraduate,

B.S. Computer Science

and Mathematics

Alex Markley

Rutgers Graduate Student,

M.S. Computer Science

Minsung Kim

Rutgers Assistant Professor, Dept. of Computer Science

Byungjun Kim

Rutgers Postdoc Researcher

3 of 29

For wireless communication, we send data using radio waves.

The path over the air is called a “channel”.

3

Set Up

4 of 29

Problem 1: Low Performance

Solution 1: MIMO

Problem 2: Noise

Solution 2: MIMO Detection

Problem 3: MIMO Detection is slow

Solution 3: ParaMax!

4

5 of 29

Problem 1: Low Performance

Solution 1: MIMO

Problem 2: Noise

Solution 2: MIMO Detection

Problem 3: MIMO Detection is slow

Solution 3: ParaMax!

5

6 of 29

Send more data at once.

We can use Multiple Input Multiple Output (MIMO) to increase throughput and service more users!

6

Solution 1: MIMO

7 of 29

MIMO Rack

7

8 of 29

8

9 of 29

Problem 1: Low Performance

Solution 1: MIMO

Problem 2: Noise

Solution 2: MIMO Detection

Problem 3: MIMO Detection is slow

Solution 3: ParaMax!

9

10 of 29

Transmitted messages are exact “symbols” (like 1s and -1s)

10

Problem

11 of 29

Noisy channel conditions can disrupt the message

What was the intended message?

11

Problem

12 of 29

Problem 1: Low Performance

Solution 1: MIMO

Problem 2: Noise

Solution 2: MIMO Detection

Problem 3: MIMO Detection is slow

Solution 3: ParaMax!

12

13 of 29

MIMO Detection maps the observed signal back to the intended message

In noisy environments, this is challenging.

13

Solution: MIMO Detection

14 of 29

Problem 1: Low Performance

Solution 1: MIMO

Problem 2: Noise

Solution 2: MIMO Detection

Problem 3: MIMO Detection is slow

Solution 3: ParaMax!

14

15 of 29

Compute distance of every combination.

The closest message is most likely to be correct.

15

Maximum Likelihood (ML)

MIMO Detection

16 of 29

Compute distance of every combination.

The closest message is most likely to be correct.

16

Maximum Likelihood (ML)

MIMO Detection

17 of 29

We want to use optimal MIMO detectors since they have lower error rate, but they’re too slow.

17

Maximum Likelihood (ML)

MIMO Detection

Time

MIMO Size

Latency Threshold

Note: Lower is better on both plots

ZF

ML

18 of 29

Problem 1: Low Performance

Solution 1: MIMO

Problem 2: Noise

Solution 2: MIMO Detection

Problem 3: MIMO Detection is slow

Solution 3: ParaMax!

18

19 of 29

Instead of trying all values.

Sample possible values and start to settle into a minima if we find one.

19

ParaMax

MIMO Detection

20 of 29

Add multiple parallel annealers, one that stays near the best solution so far and one that explores for a better minima.

20

ParaMax

MIMO Detection

21 of 29

ParaMax is designed to provide near-optimal BER and run quickly.

21

ParaMax

MIMO Detection

Time

MIMO Size

Latency Threshold

Note: Lower is better on both plots

ZF

ML

22 of 29

Our goal:

Hardware Implementation of ParaMax

First ever hardware implementation of a quantum-inspired MIMO detection algorithm

22

23 of 29

Internship Results

&

Relation to Quantum-Inspired Computing and Project Objectives

23

24 of 29

24

TIME

DATA

EMPTY

DATA

EMPTY

DATA

TX 1

TX 2

TX 3

25 of 29

25

Single Transmitter

Multiple Transmitters

Single Receiver

Multiple Receivers

Offline MIMO Detectors

Real-time MIMO Detectors

ParaMax Implementation

Blind MIMO

NOMA

26 of 29

26

We achieved perfect time synchronization, with precision down to 50 nanoseconds - 50 billionths of a second!

27 of 29

27

28 of 29

28

Single Transmitter

Multiple Transmitters

Single Receiver

Multiple Receivers

Offline MIMO Detectors

Real-time MIMO Detectors

ParaMax Implementation

Blind MIMO

NOMA (Non-Orthogonal Multiple Access)

29 of 29

Thank you!

Any Questions?

29