1. Signal Waveform Comparisons
(Rev 2)
Z. Sahinoglu, I. Guvenc, P. Orlik, S. Zhao
Digital Communications and Networking Group, MERL
Francois Chin, Sam Kwok, Zander Lei, Xiaoming Peng
Digital Wireless Dept, I2R
Tuesday, April 30, 2019

2. Option-I One Bit The Other Bit Always Empty Always Empty Always Empty
100ns
8-chip times: 150ns
100ns
8-chip times: 150ns
The Other Bit
Always Empty
Always Empty
Always Empty
8-chip times: 150ns
100ns
Enough long not to cause IFI : 100ns
8-chip times: 150ns

3. Coherent receivers exploit chip sequence patterns
Features of Option-I
Coherent receivers exploit chip sequence patterns
Non-coherent receivers see in which half the energy arrives
In ranging preamble, each piconet is assigned a different sequence of bits
Symbol rate is
1Msps (after rate ½ code)

4. Example SOP preambles in Ranging with Option-I
Piconet-I bit sequence: {1,0,0,1,1}
Piconet-II bit sequence: {1,1,0,0,0}

5. Ranging Performance of Option-I
Very resilient to SOP interference due to proper selection of preamble bit sequences
For non-coherent radios,
inter-pulse-interference due to multipath does not need to be resolved
Statistical multiplexing is needed to increase SNR due to spreading bit energy over many pulses
Preliminary results
A train of 8-pulses
EbN0 = 22dB, SIR = 0dB (randomly generated 30 symbols in the preambles of desired and interference)
CM1
Integration interval: 4ns
Ranging error: 3ns (72%)
Better accuracy if narrower energy windows

6. Modulation with Option-I (I2R’s comments )
Additional time hopping of blocks is needed to support 2-SOP
Ex: Another 160ns interval in each half of the frame [Zafer, can you illustrate this with a diagram?]
Yes Francois, I could, but I now would like revise this statement and say that “it is better if a symbol consists of multiple frames with TH of a pulse (s) in each (See below please) to avoid catastrophic collisions in the communication mode (I think this is similar to what Ismail and Patricia have in their mind).
Option-I with this additional time hopping blocks will be the only mode, since signaling is designed with SOP in mind
Option-1 will also be the same signaling scheme for synchronisation for non-coherent receiver, as common preamble are used for both sync and ranging
This will also be the common signaling scheme for the preamble in beacon packet
Coherent receiver has to perform two layer of despreading (both time-hopping and code despreading) for synchronisation
Ts <= 500ns

7. Option-II Ts = 500ns « 11 » 2-PPM + TR base M = 2 « 01 »
« 11 »
2-PPM + TR base
M = 2
One bit/symbol
« 01 »
« 10 »
« 00 »
(coherent decoding possible)

8. Option-II Features
Pulses (or doublets) are spread over the entire 500ns symbol duration
Non-coherent decoding and ranging with this waveform has not been simulated yet
Has potentially better SOP isolation than Option-I in communication mode

9. Ternary Signaling for Synchronization & Ranging
Option-III
Ternary Signaling for Synchronization & Ranging
Pulse Repetition Interval ~ 30ns
-Common signaling (Mode 1) for ALL Detectors
-Receiver-specific signaling (Mode 2) for ED 1 2 3 4 5 6 7 8 30 31
…………………………
Non-inverted pulses are blue,
Inverted pulses are green.
Synchronisation / Ranging
preamble = Binary Base
Sequence repeated
For K times…
……………
……………
Symbol Interval ~940ns
Symbol Interval ~940ns
Tuesday, April 30, 2019

10. Option-III
Criteria/Target – balance max post-despreading SNR and low auto-correlation side lobes
Ternary Seq [ ]
After Square Law
& Integration in PRI
Unipolar M-Seq [ ]
In AWGN
Soft output
Noncoherent detection of OOK
Sliding
Correlator
LPF /
integrator
BPF
( )2
ADC
Sample Rate 1/Tc
{1,-1} Binary Sequence
Bipolar M-Seq [ ]

11. Option-III Features (I2R’s comments)
Zero autocorrelation side lobe sequences (very important for leading edge detection, as the leading edge can be 6 ~ 10 dB below peak)
Under SOP interference, correlator peak is 3dB degraded in non-coherent reception due to suboptimal correlation
This makes identification of weak multipath components difficult
Suboptimal correlation sequence is chosen to achieve Zero autocorrelation side lobe, which is considered more crucial in ranging
After the square-law device, the integrator integrates over the PRI (~30ns) (simplest architecture for data comm. mode)
In option-I, integration interval is 2ns or 4ns.
Longer integration interval collects more noise
Longer integration interval collects more interference energy
After the square-law device, the integrator integrates over the 2 ~ 4ns (for ranging mode)
Same as option-I
After the square-law device, the integrator integrates over the 2 ~ 4ns (for comm. Mode with selective rake combining)
Less energy capture
Less noise capture too

12. Recommended Architecture for Ranging with Non-Coherent Rx (MERL)
(No FFT routine is needed, being different from doc#0269)
Energy image generation
Removes interference
2-4ns
Length-3 Vertical Median or Minimum Filtering
1D to 2D Conversion
LPF /
integrator
BPF
( )2
ADC
2D to 1D Conversion with Energy Combining
TOA Estimator

13. Recommended Architecture for Ranging with Non-Coherent Rx (I2R)
Energy image generation & interference suppression
2-4ns
Energy combining across symbols
Sliding Correlator
LPF /
integrator
BPF
( )2
ADC
TOA Estimator
Francois, could you elaborate on sliding correlator and energy combining blocks please, when the integration is at 4ns intervals ? . Thanks
When I tried the sliding correlator, I got results on slide 22. How would you generate your images?

14. Effect of Number of Pulses on Performance for Non-Coherent Modulation and Ranging: An Example
M: (Degree of freedom/2) = (BWxT)
N: Number of pulses
NsEb = symbol energy
No: Noise psd
Case 1: Single Pulse Per Symbol
N = 1
s = 0
s = 1 μ1
MN0
MN0+2NsEb
σ12
MN02
MN20+4N20NsEb
PDF
s = 0
s = 1
Transmitting one pulse with large energy
Energy
Energy of this pulse is NsEb μ1

15. Effect of Number of Pulses on Performance for Non-Coherent Modulation and Ranging: An Example
Case 2: Multiple Pulses Per Symbol
N = Ns
s = 0
s = 1 μ2
NsMN0
Ns(MN0+2Eb)
σ22
NsMN02
Ns(MN02+4N0Eb)
Transmitting many pulses with less energies
PDF
s = 0
s = 1
Energy
Energy of each pulse is Eb, and there are Ns number of pulses μ2

16. Effect of Number of Pulses on Performance for Non-Coherent Modulation and Ranging
The energy per symbol can be collected in a single
pulse (N=1), or in Ns pulses
The means and variances statistics of the square-law
device outputs can be observed in the absence and
presence of signal
The Euclidean distance between the means for s=0,
and s=1 are the same for both cases (2NsEb)
However, the variance term when using larger number
of pulses increases

17. Current Status
MERL is simulating Option-III according to the recommended block diagram, and will share the observations soon
Preliminary Option-I results are shared on slide 5
There are still unincorporated optimization techniques to improve edge detection performance
Adaptive threshold selection
Search back window selection etc

18. Pulse OOK vs Burst PPM (MERL Comment)
Pulse OOK (option-III)
Burst PPM (option-I)
Signaling
Spaced out pulse seq
Clustered pulse seq
Energy Integration period (for ranging)
2~4ns
Energy Integration period (for data comm.)
2ns ~ PRI (30ns)
Half Symbol period
without SOP
TBD
with SOP
Good in com. mode
(Sufficient processing gain to handle)
good in ranging mode
(Not much processing gain to handle the comm. mode)
Common signaling for preamble
No Time hopping
Time Hopping
Additional complexity for Coherent receiver to receive preamble with common signaling No
Yes (2 layer sync, TH then code de-spreading) Agreed
Inter-pulse interference during ranging operation
Less due to high PRI
33MHz PRF or ~16MHz PRF)
More due to small inter pulse interval (Yes, but IPI is not deleterious to ranging performance in burst PPM)
Sampling rate
Same in transmitter and receiver
(Isnt it different in com and ranging 4ns vs. 30ns?)
Low in receiver
High in transmitter

19. Illustrations of Energy Image Generation
To better understand how images are created, just shift the mask below over desired user and interferer signals one block at a time in the animation mode, and populate the matrix from open windows in the mask
In I2R scheme, it will be masked onto +1 and -1 positions in a symbol
Symbol duration in Option-I
Desired user code-2 {2,1,2,1}
Interference code {1,3,1,2}
User Energy matrix
Interference Energy Matrix

20. A look Into Energy Images of Option-I
EBN0 = 22dB, SIR = 0dB, no statistical multiplexing
Strong SOP interference is easily suppressed by the way the image is created and by means of length-3 minimum filtering (in ranging)
Multi-user Interference
Desired User Energy
Minimum Filtering {Length 3 Vertical}
Symbol Index
Symbol Index
Block Index
Block Index

21. Ternary Energy Image Simulation Results
Case-A: Images are generated before the Unipolar correlator
The mask is formed from positions of +1 and -1 in the sequence
Settings:
4ns block sizes
Observing the signal in 16 windows at +/- 1’s of the ternary code
Eb/N0 = 30dB
No clear sign of signal paths under AWGN + Interference, when imaging is used on data before the bipolar correlator
We will test it on samples after Unipolar correlation in the next round. I expect it to improve the images and to ease edge detection
Desired edge is here
Noise and interference free
Noise + interference (after min-3 filtering)

22. Autocorrelation of Ternary Sequences @4ns Sampling
A sequence of 4 Ternary symbols are passed through a CM1 channel (no AWGN and no SOP Interference)
Received signal energies are integrated over 4ns intervals
The sequence of energy samples are correlated with a Unipolar template (output of the correlator is plotted below)
We are studying how correlator outputs can be best exploited for ranging under AWGN and SOP interference