On the Feasibility and Performance of CDMA with Interference Cancellation

On the Feasibility and Performance of CDMA with Interference Cancellation Jack Keil Wolf QUALCOMM Incorporated Plenary Session ISSSTA 2006 August 29, 2006

Acknowledgements This presentation is based upon the work of a number of QUALCOMM personnel including: Roberto Padovani Jilei Hou Henry Pfister John Smee Joseph Soriaga Stefano Tomasin

Outline of Talk Evolution of CDMA networks From IS-95 to EV-DO and WCDMA Theoretical capacity of multiple access systems Achieving sum-rate with CDMA and interference cancellation Receiver architectures for CDMA interference cancellation Canceling pilot, overhead, and Hybrid-ARQ traffic per antenna Channel estimation techniques and performance Determining link cancellation efficiency Network simulation results for EV-DO Rev A Design tradeoffs and capacity gains from cancellation

TOPIC Evolution of CDMA networks From IS-95 to EV-DO and WCDMA Theoretical capacity of multiple access systems Achieving sum-rate with CDMA and interference cancellation Receiver architectures for CDMA interference cancellation Canceling pilot, overhead, and Hybrid-ARQ traffic per antenna Channel estimation techniques and performance Determining link cancellation efficiency Network simulation results for EV-DO Rev A Design tradeoffs and capacity gains from cancellation

Evolution of CDMA Networks The CDMA reverse link has seen wide commercial deployment due to its unique combination of: High spectral efficiency Efficient power controlled transmissions Robustness Soft handoff between cells Softer handoff between sectors of a cell Multipath diversity across the full transmission bandwidth Ease of network planning and deployment Universal frequency reuse

Evolution of CDMA Networks CDMA networks continue to evolve to support advanced applications with increased efficiency. IS-95 to cdma2000, EV-DO and WCDMA. Wide area integrated networks that support data, VoIP, video, and broadcast service with high quality of service. Extensive physical and MAC layer improvements have been combined with the classic principles of the CDMA reverse link.

Advanced CDMA Reverse Links (EV-DO, HSUPA) Each user transmits pilot, overhead, and traffic channels. The pilot is used for channel estimation. The overhead channel conveys forward and reverse link information such as: the forward link quality, the serving sector, ACK/NAK information for the forward link packets, and the reverse link data rate packet format. The traffic power governs the target traffic data rate based on the traffic-to-pilot (T2P) ratio. The T2P is controlled by an algorithm based on QoS and fairness. The traffic channel is transmitted using hybrid ARQ.

Advanced CDMA Reverse Links The total transmit power is: where E cp is pilot power. The pilot power of each user is feedback controlled. The target pilot SINR at basestation is adjusted for each user to keep the traffic channel packet error rate (PER) at 1%.

EV-DO Rev A. Reverse Link Hybrid A packet is transmitted as a series of 1 to 4 subpackets on one of 3 interlaces. Each subpacket occupies 4 slots where 1 slot = 1.66 ms = 2048 chips. The interlace structure gives the receiver time to decode and relay ACK/NAK information to the transmitter after each subpacket is received.

EV-DO Rev A. Reverse Link Hybrid ARQ

EV-DO Rev A. Reverse Link Data Rates Rate-1/5 Turbo code with repetition and puncturing is used to achieve various rates. Each entry in table has its T2P ratio chosen to achieve that target rate. The 1% PER is targeted for 4 transmissions but early termination occurs using hybrid ARQ. The modulation formats used are: BPSK, QPSK, and 8PSK.

Interference Cancellation (IC) The aim of this talk is to show how interference cancellation (IC) can improve the capacity of a CDMA system without changing the standard. We first begin with some theory that shows the benefit of interference cancellation. Then we show how IC has been implemented and the benefits achieved in an actual system.

Topic Evolution of CDMA networks From IS-95 to EV-DO and WCDMA HSPA Theoretical capacity of multiple access systems Achieving sum-rate with CDMA and interference cancellation Receiver architectures for CDMA interference cancellation Canceling pilot, overhead, and Hybrid-ARQ traffic per antenna Channel estimation techniques and performance Determining link cancellation efficiency Network simulation results for EV-DO Rev A Design tradeoffs and capacity gains from cancellation

Theoretical Limits of Multiple Access The (normalized) Shannon capacity, C, in bps/ HZ for communicating from one point to another over an AWGN channel is For any uplink multiple access system, the maximum overall throughput for K users with powers P 1, P 2, P K, is calculated by adding the received power of all users to calculate the maximal sum-rate capacity, C sum :..

Theoretical Limits of Multiple Access In CDMA without interference cancellation, each user sees the powers of all of the other users as noise. Then, the capacity for every user is given as Here, SINR is the signal-to-interference ratio and P is the sum of the powers of all of the interfering users which is modeled as white Gaussian noise. If each user has the same power P and there are K users, P = (K-1)P, and thus each user sees a capacity..

Theoretical Limits of Multiple Access Thus the sum of the capacities for the K users is:. but the maximal sum-rate capacity is:. Thus, without interference cancellation, the maximal sum rate capacity, C sum is much greater than the sum of the capacities of all of the users, KC user.

Theoretical Limits of Multiple Access In what follows, several different schemes of interference cancellation will be assumed. In each scheme, the users are decoded sequentially. This is called sequential interference cancellation (SIC). In each of these schemes, the sum of the capacities of all the users equals the maximal sum-rate capacity. In general, we will assume that there are K users, with different powers P 1, P 2, P K.

Scheme 1: Equal Powers and Unequal Rates In this scheme, we will assume that all of the users have equal powers. That is, we assume that P 1 = P 2 = = P K = P. Assume user 1 is decoded first. The decoder for user 1 sees the power from all (K-1) other users as noise. The capacity for user 1 is then:. Now assume that after decoding user 1 s signal, it is reencoded and subtracted from the received waveform before the second user is decoded. Then the decoder for user 2 sees only (K-2) other interfering signals as noise and user 2 s capacity is:.

Scheme 1: Equal Powers and Unequal Rates Now user 2 s signal is re-encoded and subtracted from the received waveform and user 3 sees a capacity. This continues until the last user, user K, sees no interference from the other users and has a capacity.

Scheme 1: Equal Powers and Unequal Rates It is an easy exercise in algebra to show that for this method of interference cancellation, C 1 + C 2 + + C K = C sum, that is, the sum of the capacities of all the users is equal to the maximal sum-rate capacity. However, in this scheme, the users have unequal capacities (or rates).

Scheme 2: Exponential Powers and Equal Rates A method of having all K users have the same capacities (or rates) is by having the users choose their powers such that the user that sees the most interference uses the most power. The interference cancellation works in the same manner as in the first scheme. That is, after each user is decoded, its information is re-encoded and subtracted from the received waveform. In particular, assume the following exponential distribution of powers. For i = 1, 2,, K, let.

Scheme 2: Exponential Powers and Equal Rates Even though the transmitted power of each user differs, it is easy to show that they have the same SINR and thus all K users have the same capacity. That is, for i=1, 2,, K,. Again, it is an easy exercise in algebra to show that the sum of the capacities for all the K users is equal to the maximal sumrate capacity:.

Scheme 2: Exponential Powers and Equal Rates However, note that this performance was achieved by having the powers vary greatly among the users. An example of this scheme is illustrated in the next slide where we have chosen K=3, N=1 and P 1 = 4, P 2 = 2, P 3 = 1.

Scheme 2: Exponential Powers and Equal Rates Decode User 1 first with Users 2,3 as noise. SINR 1 = 4/(2+1+1)=1 with capacity of log 2 (1+1) = 1 bps/hz. Subtract User 1, so User 2 sees only User 3 as noise. SINR 2 =2/(1+1)=1. Capacity again equals log 2 (1+1) = 1 bps/hz. Subtract User 2. User 3 sees only noise. SINR 3 = 1/1. Capacity equals log 2 (1+1) = 1 bps/hz. Summing each user s capacity gives 3 bps/hz. But as shown below, the maximal sum-rate capacity equals 3 bps/hz..

Scheme 1 : A Modification of Scheme 1 Remember that in Scheme 1 all users have the same power but different capacities. It is easy to modify it so that the K users have both equal powers and also equal capacities (on the average). The modification is as follows: In the 1 st transmission decode in the order 1, 2,, K-1, K. In the 2 nd transmission decode in the order 2, 3,, K, 1 In the 3 rd transmission decode in the order 3, 4,, 1, 2 In the K th transmission, decode in the order K, 1,, K-2, K-1 The average capacity of all K users is now the same and their sum is equal to the maximal sum-rate capacity. Furthermore, all K users have the same power.

Scheme 3: Asynchronous CDMA. Equal Power and Equal Rates We now show another method with equal powers and equal capacities that is more applicable to CDMA 2000. In this scheme, the users packets arrive at different times. Apply sequential IC by decoding and removing each packet when it arrives. Each packet sees the interference from future un-decoded packets (but not past decoded packets) from other users. This is illustrated on the next slide for the case of 3 users.

Scheme 3: Asynchronous CDMA. Equal Power and Equal Rates Divide frame of User 1 into 3 sub-frames: (i), (ii), and (iii). Each sub-frame sees a different noise level of 1, 7/3+1, and 2x7/3+1. This is time-multiplexing of 3 parallel channels with different SINR s and thus different capacities. We average the capacity of the parallel channels to find that each user s capacity is 1bps/Hz (see below). The sum of the capacities equals the maximal sum-rate capacity:

Other Schemes for Achieving the Maximal Sum-Rate Capacity By SIC Combinations of the previously described schemes are possible that also achieve the maximal sum-rate capacity. Also generalizations of these schemes exist such as schemes with: Non-uniform time offsets Other than exponential power distributions Different user data rates

Robustness of Sequential Interference Cancellation In scheme 2, frame synchronous interference cancellation with equal rates but unequal powers, the last user to be decoded is received at low power and its successful decoding is highly dependent on the successful decoding of prior users. This is not a robust scheme! Scheme 3. frame asynchronous interference cancellation with equal rates and equal powers, is a much more robust scheme since all users are treated equally. The hybrid-arq used in CDMA 2000 is even more robust since it is frame asynchronous and there are multiple decoding attempts for each packet.

Applying SIC Principles to Commercial In CDMA networks, different users have different transmission rates and different received powers. CDMA networks intentionally stagger user frame offsets in time. It makes decoding and backhaul usage more uniform in time. In EV-DO RevA, 4 frame offsets are used That is why a subpacket is said to take 4 slots.

Moore s Law Technology Advances Advances in digital processing technology now allows the robustness of CDMA to be combined with capacity approaching techniques of SIC. QUALCOMM s CSM6800 chipset solution presently implements pilot interference cancellation for 3G CDMA to increase the EV- DO VoIP capacity by about 10-15% more users. QUALCOMM is now developing CDMA products that incorporate pilot, overhead, and traffic interference cancellation.

Canceling Pilot, Overhead, and Traffic Interference cancellation can be implemented by subtracting signals from the CDMA basestation buffer once those signals are no longer useful. Cancel pilot after channel estimation no decoding is required. Cancel overhead after overhead channel decoding. Cancel traffic after data channel decoding. Subtractive IC can be combined with other interference suppression techniques. For example, it can be combined with various spatial interference suppression techniques using multiple receiver antennas such as null steering.

CDMA IC Receiver Architecture With more than one receiver antenna, a capacity achieving approach is a per user MMSE front-end followed by interference cancellation. Interference is reconstructed and subtracted from each antenna based on multipath channel estimates. In addition, one can use re-encoded data to better estimate the channel.

Principles of CDMA IC with H-ARQ and Asynchronism We use closed loop power control same as today. Sequential interference cancellation approaches the maximal sum-rate capacity without any need for any special power shaping among users. The channel fading governs the decoding order. After a packet decodes, we reconstruct the transmitted stream and convolve it with the channel estimate. We then subtract it in the receiver buffer. The receiver buffer spans the packet length. For EV-DO RevA, a packet with all 4 subpackets spans 40 slots. We demodulate the packets based on samples with the pilot, overhead and decoded packets removed.

Principles of CDMA IC with H-ARQ and Asynchronism On the average, ¼ of the active users finish their subpackets in each of the 4 slots. With Hybrid-ARQ, we demodulate and attempt to decode the user s traffic after each subpacket is received. We demodulate the latest arriving subpacket and combine it with any previous subpackets stored in the receiver buffer. After successful decoding, all subpackets stored in the buffer that were involved in that decoding are remodulated and convolved with the channel estimates for interference cancellation.

Receiver Buffer and Iterating IC between Frame Offsets

Receiver Buffer and Iterating IC between Frame Offsets To keep up with real time, receiver must process one frame offset worth of users in 1 slot of real time. As we shall see, sometimes the processor is faster than that. After finishing demod/decode of the users in the current frame offset, receiver can iterate back and reattempt demod/decode on previous slots.

Successive and Group Interference With successive IC, a given user is demodulated, decoded, reconstructed, and subtracted before the next user is demodulated. It is optimal if the best decoding order is known a-priori. It is inefficient because there is no pipeline. Other receiver stages wait idle as a single user moves through each stage. The other extreme is group IC where users in the same group are processed in parallel. Users do not see mutual IC gain at the current decoding attempt. We group users by frame offset. Users finishing subpackets at same time can see mutual IC gain when future subpackets of the un-decoded user arrive.

Successive and Group Interference Cancellation A realistic pipeline has performance between Group IC and Sequential IC. Iterating Sequential IC with random decoding order and iterating Group IC allow both to converge to performance of Sequential IC with optimal decoding order.

Example of H-ARQ and Cancellation Buffer Depiction of 1 interlace in buffer as successive subpackets arrive and decoded packets get cancelled.

Average Remaining Power in Cancellation Buffer Since subpackets are continually being cancelled as packets decode there is less interference deeper into the buffer. ROT is Rise Over Thermal which is defined as (signal + interference + noise) / noise.

Example of Subpacket SINR Gains from IC CDF of effective SINR of the 1 st subpacket at the time at which we attempt to decode the packet when 1, 2, 3, 4 subpackets arrived, conditioned on the packet decoding on the 4 th subpacket. The subpacket SINR gets better as it ages through the buffer.

Cancellation Factor and Residual Power In interference cancellation, we don t attempt to subtract out all of the interference. Rather, we subtract a scaled version of the reconstructed signal where the cancellation factor is chosen to minimize the residual power. The optimal cancellation factor and the residual power are plotted on the next slide as a function of the signal to noise ratio for the case of no fading.

Cancellation Factor and Residual Power (No Fading) Use larger scaling factor at higher SINR since we are more confident of the estimate. (N is processing gain.)

Link PER Curves with Bit Exact Cancellation Computing the bit PER curve with and without IC is the same as applying an effective SINR for IC based on cancelled powers. The decoding performance at the link level is well characterized by the average SINR input to the decoder, regardless of whether the noise is interference from other users or residual signals after cancellation. β is the fraction of power cancelled for each signal on each antenna.

Network Simulation Parameters Based on the 3GPP2 EV-DO evaluation methodology. 19 cells with 3 sectors/cell for 57 sectors with 10 users/sector. Wrapped around to remove edge effects for users in outer cells. Each basestation has 2 or 4 antennas per sector. 2 km between adjacent basestations, path loss exponent 3.5. Zero mean lognormal shadowing with σ=8.9db Drop users if total path loss is above 138.5dB

Standard Mix of Channels CDMA The standards specify channel models to apply to the users. 3GPP2 Name Probability in Standard Mix Fading Speed Delay Profile Channel A 30 % 3 km/hr Pedestrian A Channel B 30 % 10 km/hr Pedestrian B Channel C 20 % 30 km/hr Vehicular A Channel D 10 % 120 km/ hr Pedestrian A Channel E 10 % 1.5 Hz Ricean, The details of these channel models are omitted. 10dB

Network Simulation Modeling Each sector computes the average rise-over-thermal (RoT) across its antennas for every slot Each sector transmits a reverse activity bit (RAB) in each slot to control the T2P. RAB = 1 if RoT > threshold RAB = 0 if RoT < threshold E.g. threshold = 5.8 db for 2 antennas, 6.5 db for 4 antennas

Network Simulation Modeling Assumes full buffer users with T2P set for 1% PER after 4 subpackets Users increase T2P if RAB = 0 Users decrease T2P if RAB = 1 With IC, one should use effective RoT (calculated post-ic) to more accurately reflect the interference level. IC allows more power to arrive at the basestation. The result is higher data rates in the traffic channels.

Distribution of RoT and Pilot SINR Overhead channels are decoded pre-ic with larger interference. Shift all T2P ratios for IC system down, as accommodated by EV-DO. Power control maintains 1% data PER by increasing pilot power. Brings pilot SINR up to preserve overhead channel performance.

Memory Size for Front-End IC Buffer With H-ARQ can only re-demodulate subpackets stored in buffer. Diminishing IC gain with increased buffer size since many packets decode after 2-3 subpackets. 40 slot buffer design allows uniform treatment of all subpackets. 3 sectors x 4 antennas x 40 slots x 2048 chips/slot x 2 samples/chip x 16 bits/sample = 30 Mbits Can now be efficiently implemented in single transistor embedded DRAM technology.

Iterating Between Users on Previous Frame Offsets IN EV-DO Rev A, there are 4 possible frame offsets, i.e., 4 slots per sub-packet. In one slot (1.67 ms), we have to process all the users in that frame offset. But the processor may be fast enough to do more than that. We can use this extra processing power to process users in other frame offsets. In some cases, we have enough time to iterate between frame offsets. For example, we might process: n, n-1, n-2, and then n, n-1, n-2.

Iterating Between Users on Previous Frame Offsets In 1 slot real time, receiver can iterate IC between frame offset groups. Can see convergence between group IC and SIC due to iterations. More gain for processing slots n, n-1 than processing slot n twice. More gain for processing n, n-1, n-2 than n, n-1, n, n-1. With overall increase from 58% to 78%, note 71% gain for n, n-1 alone. Throughput Gain over No IC (%) Group IC 2-Iter. Group IC SIC 2-Iter. SIC Slot n 58.4 % 65.9 % 62.8 % 66.4 % Slots n, n-1 Slots n, n-1, n-2 71.0 % 75.5% 73.4 % 77.5% 72.6 % 76.7% 73.4 % 77.5%

Comparison of Channel Estimation Techniques On the next slide we give results for 3 types of channel estimation techniques for both 2 and 4 antennas. They are: Pilot Channel Estimation (Pilot CE) where only the pilot is used in the estimate. Data Parallel Channel Estimation (Data PCE) where all Rake fingers are estimated in parallel using re-encoded data. Data Iterative Channel Estimation (Data ICE) where the Rake fingers are estimated iteratively using re-encoded data.

Comparison of Channel Estimation Techniques Data based estimation gives good gain over pilot based estimation. Refining data estimation based on ICE gives only small gain over PCE. Good tradeoff is to estimate all fingers of a user in parallel based on reencoded data.

Throughput Distribution Across Users Absolute throughput shows gains across all rates. Relative throughput (each user s average rate normalized by the sector throughput) is unchanged by IC fairness is preserved.

Conclusions Evolution of CDMA networks Physical and MAC layers of EV-DO and WCDMA are evolving to support more applications with higher efficiency. Theoretical capacity of multiple access systems Can apply principle of achieving sum-rate capacity with CDMA and SIC to systems with multiple receiver antennas, asynchronism, and H-ARQ. Receiver architectures for CDMA interference cancellation Can be commercialized with current technology and incorporated into the network without modification to the system operation. Channel estimation techniques and performance Can use re-encoded data and CDMA processing gain to achieve high cancellation efficiency on multipath fading channels. Network simulation results for EV-DO Rev A Practical receivers that span all subpackets and iterate IC over several frame offsets achieve throughput gains over 60% across users.

Some References 1. J. Hou, J. E. Smee, H. D. Pfister, S. Tomasin, Implementing Interference Cancellation to Increase the EV-DO Rev A Reverse Link Capacity, IEEE Comm. Magazine, pp. 96-102, Feb. 2006. 2. N. Bhushan et. al., cdma2000 1xEV-DO Revision A: A Physical and MAC Layer Overview, IEEE Comm. Magazine, Feb. 2006, pp. 75-87. 3. A. J. Viterbi, Very Low Rate Convolutional Codes for Maximum Theoretical Performance of Spread-Spectrum Multiple-Access Channels, IEEE Journal on Selected Areas in Communications, JSAC Vol. 8, No.4., May 1990, pp. 641-649. 4. S. Verdu, Multiuser Detection, Cambridge University Press, 1998. 5. R. G. Gallager, Information Theory and Reliable Communications, John Wiley & Sons, 1968. 6. D. Tse and P. Viswanath, Fundamentals of Wireless Communication, Cambridge University Press, 2005.