andrea zanella, andrea biral, michele zorzi {zanella, biraland, zorzi}@dei.unipd.it university of...
TRANSCRIPT
Andrea Zanella, Andrea Biral, Michele Zorzi
{zanella, biraland, zorzi}@dei.unipd.it
University of Padova (ITALY)
Asymptotic Throughput Analysis of Massive M2M Access
Outline
The challenge of massive M2M access Random access with MPR and SIC Approximate throughput model Asymptotic analysis Conclusions
ITA - 2015 2
Challenges for M2M access
Massive number of users
Sporadic traffic
Short messages
Current access schemes are not adequate
for this type of scenario Costly first access mechanisms
Lack of effective ways for massive access
ITA - 2015 3
Techniques for improved access
Capture phenomenon Successful reception in the event of a collision Many models exist, based on power/time of
arrival/distance relationships, number of overlapping signals/etc.
Many papers in the literature
Multi-Packet Reception capability The ability of a receiver to decode multiple overlapping
packets Requires some advanced PHY technique (CDMA, MIMO,
IC, etc.)4
Massive asynchronous access
Approach move complexity to BS use advanced MAC/PHY
MPR: multi packet reception SIC: successive interference cancellation
Some relevant questions: How many transmitters can be served? What is the maximum cell throughput? How can it be achieved?
ITA - 2015 5
Physical capture model
j > b j-th signal is correctly decoded (capture)
j <= b j-th signal is collided (missed)
Aggregate
interference
ITA - 2015
TX1
TX2
TX3
TXj
TXn
Pj
PnP1P2
P3
RX
Performance analysis
Number of simultaneous transmissions (n)
Statistical distribution of the received signal powers (Pi)
Capture threshold (b) Max number of SIC
iterations (K) Interference
cancellation ratio (z)
System parameters
7ITA - 2015
Performance analysis
Capture probability
Cn(r;K)=Pr[r signals out of n are captured within at most K SIC cycles]
Computing Cn(r;K) is difficult because the SINRs are all coupled
E.g.
Computation of Cn(r;k) becomes more and more complex as the number n of signals increases
SIC makes things even more complex
8ITA - 2015
Computation of capture probs
Narrowband (b>1), No SIC (K=0) [Zorzi&Rao,JSAC1994,TVT1997] derive the probability Cn(1;0) that one signal is captured
MPR and SIC are not considered
Wideband (b<1), No SIC (K=0) [Nguyen&Ephremides&Wieselthier,ISIT06, ISIT07] derive the probability 1-Cn(0;0) that at
least one signal is captured Expression involves n folded integrals, does not scale with n
Wideband (b<1)+SIC (K>0) [ViterbiJSAC90] shows that SIC can achieve Shannon capacity in AWGN channels
Requires suitable received signal power allocation
[Narasimhan, ISIT07] studies outage rate regions in presence of Rayleigh fading Eqs can be computed only for few users
[Weber et al, TIT07] study SIC in ad hoc wireless networks Derive bounds on the transmission capacity based on stochastic geometry arguments
[ZanellaZorzi, TCOM2012] provide a scalable method for the numerical evaluation of the capture probability distribution Cn(r;K), and simple approximate expressions
9ITA - 2015
Approximate mean number of captures: first
reception Iteration h=0: number of undecoded signals n0=n
decoded signals, with mean
Approx capture threshold
Approx capture condition
Mean number of decoded signals
Mean number of still undecoded signals
PEnbI 10
10
Approximate mean number of captures: h-th
iteration Iteration h>0: avg number of undecoded signals:
Approximate capture threshold
Approximate capture condition
Mean number of decoded signals
Mean number of still undecoded signals and average throughput
1Pr~ hhhh IPIPnr
)()|( 11 hPhhPhh InFIPIFnn
•Residual interf.•Interf. from undecoded signals
11
1)1( hn nnhS
SIC+MPR throughput
b=0.02
Rayleigh fading
# of SIC iterations
High congestionLow congestion
optimal # of concurrent
transmissions
ITA - 2015 12
ApproxSimulation
Fixed point throughput approx.
Letting # of SIC cycles go to infinity, the residual interference can either go to zero all signals are eventually
decoded and the throughput equals the number n of overlapping transmissions
or reach a steady value I∞(n) which is the fixed-point solution of the equation:
Average throughput in the limit:
ITA - 2015 13
Approx asymptotic throughput
Throughput grows linearly with n until the equation returns non-zero solution(s) x>0
Max throughput equals where n* is the value of n for which x is minimized
To find n*, we rewrite the eq. as:
ITA - 2015 14
Approx asymptotic throughput
We can also prove that n* is the optimal number of transmissions, i.e.,
In fact:
Which is true since
ITA - 2015 16
Asymptotic performance
ITA - 2015 17
Analytical throughput estimate is reasonably good for small values of b• Analysis is accurate in the
range of interest (massive low-rate access)
Optimal throughput scales linearly with 1/b• It is possible to serve
twice as many users at half the rate
• An arbitrarily large number of nodes can be served (but check OH)
Conclusions We proposed an approximate analysis of the
asymptotic throughput of random wireless systems with MPR + SIC
The mathematical model is shown to be slightly optimistic in estimating the throughput, but it captures correctly the fundamental behaviors With ideal SIC, MPR capabilities can be fully exploited
even using a simple slotted random access mechanism Achieving the optimal performance requires an accurate
control of the total number of transmitters Throughput grows almost linearly with 1/b
ITA - 2015 18
Future work Improve the accuracy of the mathematical model for large
values of SIC iterations Some ideas in the paper
Relax some simplifying assumptions, such as ideal SIC Account for residual interference
Include protocol aspects into the model How to control access in a decentralized fashion
Investigate energy aspects Very sensitive in M2M scenarios
ITA - 2015 19