E N D
1. Channel Aware Random Access in Wireless Networks Professor Ye (Geoffrey) Li
School of Electrical and Computer Engineering
Georgia Institute of Technology
Atlanta, GA, 30332-0250
Collaborated by G. Ganesan
2. Outline Introduction
CAMCRA scheme
Imperfect CSI
Stability
Summary
3. Outline Introduction
CAMCRA scheme
Imperfect CSI
Stability
Summary
4. Random Access - Illustration
5. Random Access Schemes Features:
No centralized scheduling
Efficient for bursty traffic
Simple to implement
Data rate ? governed by packet success probability
Requirements:
High uplink data rate
Low packet delay
Our Work:
Exploit multicarrier and multi-user diversity for random access!
6. Outline Introduction
CAMCRA scheme
Imperfect CSI
Stability
Summary
7. System description Aloha ? template random access scheme
Multicarrier networks only
Rayleigh fading channel
Capacity achieving codes
8. Aloha Most popular random access scheme.
Basic idea : Transmit whenever required
Simple but very powerful
9. Collision Model
One packet sent ? succ. received
More than one packet sent ? all packets lost
Stricter model
Tractable analysis
10. Channel Aware Aloha (CAA) MAC layer has complete knowledge of present channel conditions (Qin03)
CAA scheme for n users
Optimal H0 = H0(n) chosen s.t.
Max. prob. of success
11. Example – user network
SNR of the ith user ?
Rayleigh fading ?
Threshold of the ith user ?
Channel access probability for ith user ?
Equal access probability for “unequal” users!
CAA applicable to non-homogenous networks!
12. Asymptotic Throughput Defined as
? throughput when network pop. is
For constant rate network
? success prob.
For variable rate network
? actual bit rate
13. Asymptotic Throughput N user network
Assume total input process ? Poisson( )
Assume each user can hold arbitrarily large packets
Theorem (Bert00): For stability of network, max. allowable input rate satisfies
14. Aloha Success Probability For Channel Aware Aloha,
Max. allowable input rate =
15. Variable Rate Networks Assume adaptive modulation
Assume presence of capacity achieving codes
Then possible to increase throughput in Channel Aware Aloha
How? Exploit Multiuser Diversity
16. Multiuser Diversity Assume users with channel gains i.i.d.
Capacity when one randomly chosen user transmits
(bpcu: bits per channel use)
Capacity when user with best gain transmits
for large N
17. Throughput of CAA Max. Achievable throughput:
prob. of success × capacity ?
In CAA, user with best gain always transmits
Hence,
18. Novel CAMCRA scheme CAMCRA: Channel Aware Multi-Channel Random Access (Ganes05)
Features:
Exploits multicarrier diversity…
Offers throughput gain…
Robust to channel uncertainties…
20. CAMCRA scheme ? existing random access scheme for users (e.g. channel aware Aloha)
?gain of th user at th carrier
For each user , , choose carriers where
?number of users assigned to th carrier
Apply, on carrier , the scheme
21. Example : CAMCRA
22. Probability of collision is reduced.
Without CAMCRA :
ave. # of users per carrier =
With CAMCRA :
ave. # of users per carrier =
Very low chances of collision if high!!
23. Camcra - features How about if each user selects one carrier randomly
? multicarrier diversity lost
? provides for constant rate networks
? gain only in success probability
24. Constant rate CAMCRA
25. Camcra - features Rem: In CAMCRA
? each user selects best carrier among carriers
? gain of best carrier ˜ (multicarrier diversity)
? variable rate networks
26. Variable rate CAMCRA
27. Outline Introduction
CAMCRA scheme
Imperfect CSI
Stability
Summary
28. Imperfect CSI So far…assumed perfect channel conditions and statistics…
In reality…channel estimate is imperfect…
Actual ? channel gain:
variance:
Estimated ? channel gain:
variance:
29. Modelling Imperfect CSI
30. Channel Uncertainty Assume soft estimator…
, ?complex Gaussian RV’s (poss. dep.)
Define uncertainty
? perfect CSI
? no knowledge of CSI
In practice, or
31. Imperfect CSI in CAA
CAA( ) ? CAA scheme under uncertainty
? probability of successful transmission under CAA( ) scheme
When , (perfect CSI), already seen…
When , we have
32. Surprising Result!! For , already know that
= Max. Allowable input rate
as
However for any , we have
= Max. Allowable input rate
as !!!
33. CSI uncertainty in CAa
34. More Surprising Rem: Max. Achievable throughput
= prob. of success × capacity
When ,
When ,
as
35. Consequences Any uncertainty in CSI in CAA
zero asymptotic throughput
zero maximum allowable input rate
Questions:
Can this be remedied?
How does CAMCRA perform?
36. How About CAMCRA? More robust to imperfect CSI?
Non-zero asymptotic success probability?
37. Interesting fact (Guocong05)1: OFDM with carriers with perfect CSI ( )
max. capacity =
When , obviously…
max. capacity <
In fact,
max. capacity =
where
and ? estimation errors with
38. lower bound Theorem (Ganes06): for any ,
where
.
Capacity retains the same order!
However, the bound is loose!
39. Interpretation ? extra SNR needed to offset channel
uncertainty
Multicarrier network ? very robust to imperfect CSI in downlink direction
Throughput utmost reduced by a constant factor…i.e.
max. capacity =
40. CAMCRA Throughput
Theorem (Ganes06): In CAMCRA, when
as
Throughput reduced utmost by constant factor!!
Rem. in CAA: When ,
as !
41. CAA vs CAMCRA (? = 10%)
42. CAA vs CAMCRA (? = 50%)
43. Robustness of CAMCRA to channel imperfections
44. Outline Introduction
CAMCRA scheme
Imperfect CSI
Stability
Summary
45. Two user aloha -assumptions Two users U1 and U2
Total N carriers
Total Bandwidth = B Hz
ai ? fraction of carriers selected by Ui
Rayleigh fading, independent across carriers
Stability Region: set of allowable input rate vectors s.t. the lengths of queues are finite
46. Major results CASCRA1 ? If Ui chooses ai of his carriers randomly
Net (normalized) throughput = ai
CAMCRA2 ? If Ui chooses ai of his best carriers
Net (normalized) throughput = C(ai)
? Later we show, C(ai) > ai
? Leads to enlarged stability region!
47. Assumptions for Throughput analysis adaptive modulation
capacity achieving codes
48. Throughput analysis (contd) When Ui selects ai of his carriers randomly…(CASCRA)
? Max. possible throughput = ai
? Max. allowable input rate = ai
Normalized input rate : = ai
49. theorem (Ganes07): When Ui selects ai of his best carriers …(CAMCRA)
? Max. (normalized) throughput = C(ai)
? Max. (normalized) input rate = C(ai)
where
50. Throughput comparison
51. observations C(a) : concave function of a
We have C(a) > a
For same bandwidth, CAMCRA can support higher data rate than CASCRA
Can this lead to larger stability region?
52. Assumptions for Stability analysis Assume buffer can hold arbitrarily large number of packets
Assume input arrival rate ? Poisson process
Input arrival process independent from user to user
53. Two user aloha Stability region (SALOHA): Set of all input rate vectors “supported” by Aloha
“Supported” ? Loosely implies buffer of each user does not overflow
Two user system: SALOHA ? well known (single carrier)
54. Stability region (CASCRA vs. CAMCRA) , ? normalized input rates for two user systems…
CASCRA stability region (Tysb78) :
CAMCRA
Exploits presence of multiple carriers
Closed form expression not compact
Larger stability region!!
55. Stability region
56. Outline Introduction
CAMCRA scheme
Imperfect CSI
Stability
Summary
57. Summary (camcra) Multicarrier diversity useful in random access schemes
Probability of collision reduced ? throughput improvement
Throughput improvement by 50% in Channel Aware Aloha
58. Summary (imperfect CSI)
Studied effect of imperfect CSI on CAA and CAMCRA…
Showed ? CAA highly sensitive to CSI imperfection…
Asymptotic throughput ? 0 in CAA under imperfect CSI…
However ? CAMCRA very robust to imperfection in CSI
Asymptotic throughput utmost reduced by a constant factor!!
59. Summary (stability region) Studied effect of multicarrier (m/c) diversity…
Derived throughput expressions for two user networks with m/c diversity
Characterized stability region of two user networks with m/c diversity
Showed enlargement of stability region using multiple carriers
60. References G. Ghurumuruhan, Y. (G.) Li, and F. W. Vook, “Stability region of multicarrier channel aware Aloha,” Proc. of IEEE Wireless Commun. and Networking Conf., Hong Kong, China, March 2007 / also to appear in IEEE Trans. Inf. The., June 2007.
G. Ghurumuruhan, Y. (G.) Li, and A. Swami, “Channel aware Aloha with imperfect CSI,” Proc. of IEEE 2006 Global Telecommun. Conf., San Francisco, CA Nov. 2006 /also submitted to IEEE Trans. Inf. The.
G. Ghurumuruhan, G.-C. Song and Y. (G.) Li, “Asymptotic throughput analysis of distributed multichannel random access schemes,” Proc. of 2005 IEEE International Conf. on Commun., pp. 3637-3641, Seoul, South Korea, May 2005 //also submitted to IEEE Trans. Inf. The.