HSDPA - High Speed Downlink Packet Access

Transcription

1 GROUP 3 1 HSDPA - High Speed Downlink Packet Access Evaluating HARQ with Soft Combining for HSDPA Daniel Arenhage, Joel Pettersson, Peyman Barazandeh, Andre Laszlo MPNET, MPCOM, MPNET, MPNET {arenhage, petjoe, peymanb, andrel}@student.chalmers.se Abstract 3G (Third Generation) cellular networks provide high speed data packet transfer both at uplink and downlink using High Speed Packet Access (HSPA) technology. High Speed Downlink Packet Access (HSDPA) as the 3G evolution of Wideband Code Division Multiple Access (WCDMA) enjoys a couple of key-enabling features such as fast link adaptation, fast scheduling and Hybrid Automatic Repeat-reQuest (HARQ) with soft combining in order to provide high speed data packet transfers at downlink. In this paper, we introduce some of the fundamental technologies of HSDPA, and then focus on HARQ with soft combining as one of the remarkable features of HSDPA. HARQ and HARQ with soft combining as well as its two different strategies, Chase combining and Incremental Redundancy (IR), are described. We evaluate these two flavors of HARQ with soft combining and explain their pros and cons through comparing them. The simulations aim to investigate how a channel with the HARQ of HSDPA, but without the link adaptation will perform compared to a regular HSDPA channel. The simulations also provide a comparison between applying redundancy in terms of error correction and error detection schemes versus transmitting simple signals without any redundancy concerning error correction. Index Terms HSDPA (High Speed Downlink Packet Access), ARQ (Automatic Repeat-reQuest), HARQ (Hybrid Automatic Repeat-reQuest), Chase combining, Incremental redundancy, wireless data networks, simulation I. INTRODUCTION The performance of 3G and mobile communication has increased dramatically within the past couple of years. The pressure of today s mobile users has forced manufacturers to push the limits in speed and quality in order to be able to compete in the market of mobile communication. In the transaction from Second Generation-2G (GSM, EDGE) to 3G, there has been a significant change in speed. The first 3G, also known as WCDMA (Wideband Code Division Multiple Access) was introduced in the late 1990s and was fairly quickly accepted as a standard in the family of 3G technologies. With the rapid development in mobile industry and the constant pressure of increasing speed, further upgrades were needed. The upgrade of the current 3G technology is known as High Speed Packet Access (HSPA). HSPA is a collective name of the two different protocols, High Speed Downlink Packet Access (HSDPA) and High Speed Uplink Packet Access (HSUPA). This paper focuses on the downlink part of HSPA and will briefly go through some of the new techniques applied in order to increase the downlink speed of the first 3G. Later in this report we will go deeper into one specific topic called ARQ that introduces some interesting features towards higher rates in downlink.[1] HSDPA is a collective name of different techniques applied to increase the downlink speed in mobile broadband. Mainly the first version of HSDPA were only a software patch that was able to increase the speed from WCDMAs theoretical speed of 2Mbps to HSDPAs 14.4 Mbps. The basic principle of HSDPA is based on using all available resources in every cell not currently used to provide service to one or more users.[1] High Speed Downlink Shared Channel (HS-DSCH) was introduced in order to be able to provide the new high data rates in HSDPA. The HS-DSCH code resource is shared, intracell orthogonally, primarily in the time domain, but also in the code domain, and makes use of a spreading factor of 16. The channel is divided in the time domain into 2.01 ms intervals called transmission time intervals (TTI). In the simplest case, a user is assigned a whole TTI but if the whole resource is not needed, further sharing can be achieved in the codedomain by assigning different channelization codes to different users. HSDPA leaves it up to the base station designer to implement the resource scheduler in HS- DSCH, but HSDPA allows the resource scheduler to take both channel conditions and traffic conditions into account. The short TTI of HSDPA also allows the scheduler to track fast channel variations. A longer TTI would make the delays longer, and a shorter TTI would increase the overhead. [1] [6] Since the HS-DSCH is rate-controlled rather than powercontrolled, the HSDPA specification indicates that all remaining power should be allocated to HS-DSCH after other power-controlled services have been served. This makes the used cell power used in a given instant relatively constant. To compensate, HS-DSCH selects a data-rate based on the channel conditions and the amount of power available. The channelization codes of HS-DSCH are known as high-speed physical down-link shared channels, or HS-PDSCH. Between 1 and 15 codes can be allocated to HS-DSCH, this allows the operator to configure for a trade-off between HS-DSCH and other channels, such as circuit- switched services and control signaling. The first node in the code tree, however, can never be used for HS-DSCH. It is used for mandatory physical channels, such as the common pilot channel. [1] [6] In addition to the software updates applied, a new architecture reducing the load on earlier overloaded components got available. The architecture for HSPA is called Release 99 UTRAN. (See fig.1) With some of the new techniques available developers were able to move some of the functionality earlier provided by the Radio Network Controller (RNC) to be implemented in the so called nodeb instead. With this new sub-merge of

2 GROUP 3 2 and requests retransmission in the case that an erroneous signal is received. In standard ARQ, an error detection scheme such as Cyclic Redundancy Check (CRC) is used, and redundant bits are added to the original transmitted data [1, p. 118]. Fig. 1. UTRAN architecture [1] techniques a new entity was needed in order to transmit data on the HS-DSCH between User Equipment (UE) and UTRAN side. This new entity was introduced in form of a new MAC sub-layer called MAC-hs [1, c.9] [3]. MAC-hs was intended to be responsible for e.g scheduling, formatting, rate control and some ARQ operations. MAC-hs was introduced as a new entity both on the UTRAN side as well as on the UE side to handle both allocated physical resources and data transmissions towards the HS- DSCH. The data is processed within the MAC-hs and passes via the HS-DSCH channel to the HS-DSCH physical layer. In the physical layer processing, the transport block receives some extra attachments such as Cyclic Redundancy Check (CRC) needed by the UE, and goes through a series of modulation and encoding, preparing it for transmission. In order to support reordering and demultiplexing of the incoming MAC-d Protocol Data Units (PDUs), MAC-hs carries an extra header that contains relevant information [1, c.9] [3]. The 3G network today using HSPA standards applies techniques to deal with the variations that occur in the channel condition between mobile units. Due to rapid variations and interference in the channels, different techniques are needed to be applied in order to sustain a good quality in transmissions. The earlier versions of 3G were based on dynamic transmitpower control to deal with the variations in channel conditions. This technique is based on collecting data from previous transmissions in order to adjust upcoming transmissions, and varying the transmit power accordingly. In HSDPA, dynamic rate control was introduced to deal with this issue. Dynamic rate control is a combination of dynamically changing the modulation scheme and the channel-coding rate in order to match the instant changes in radio condition. In comparison to transmit-power control, with rate control you could increase the data-rate significantly and let the user benefit from higher bandwidths [1, c.7, c.9]. The rest of the report will focus on understanding and comparing different techniques applied in Automatic RepeatreQuest (ARQ). A. ARQ II. TECHNICAL STUDY Automatic Repeat-reQuest (ARQ) is a widely used technique in wireless mobile networks to detect transmission errors, and it implies that the receiving end detects frame errors B. Hybrid ARQ Channel dependent scheduling and link adaptation are two main features of HSDPA among its other features and they are achieved through proper data processing before data transmission in order to exploit wireless channel variations. But, radio-link quality is susceptible to random variations and it is impossible to perfectly adapt to the instantaneous radiolink variations, therefore it is very common to receive data packets with errors. Hybrid ARQ, also known as HARQ, is a very efficient technique that employs a coding scheme through which some frame errors can be corrected [1, p. 105]. The advantage is reduction in the average number of transmissions although each transmission carries redundant information. In HARQ, a subset of all errors is corrected using Forward Error Correction (FEC) bits which are redundantly added to the existing Error Detection (ED) bits. In poor signal conditions HARQ outperforms ARQ but in good signal conditions HARQ has noticeable performance decrease. Basically, HARQ falls into two categories: Type I and Type II. Type I is the simplest form in which sender encodes the message through adding both ED and FEC information before transmission. Receiver, first decodes the error-correction code to determine channel quality, and decides whether retransmission is needed based on the channel quality. In the case that the channel quality is acceptable enough, all transmission errors are correctable and the received data block can be obtained error free. When channel does not have proper quality, the received coded data block is rejected, and retransmission request is sent to the sender. Type II applies more complex operation than Type I. Sender sends the original message plus ED codes (and no FEC information) if the first transmission is received error free, and in the occasion that an error is detected by receiver FEC is added to the sent data along with the preexisting ED. It is also possible to combine two consecutive error-free transmissions when none of the two sent messages has error [1, p ]. C. Hybrid ARQ with Soft Combining One of the main features of HSDPA is utilizing Hybrid ARQ with soft combining. In HARQ, the erroneously received data packet is thoroughly discarded, and retransmission request is sent to sender. In spite of the fact that the received data packet cannot be decoded, the received signal still contains information, and discarding the whole received packet means losing information. Hybrid ARQ with soft combining handles this drawback. The basic notion is to store erroneously received packets in a buffer memory at receiver side instead of discarding them, and after receiving correct retransmitted packet, incorrect buffered data block is combined with error free data packet to obtain a more reliable single packet via providing higher probability to increase successful decoding [1, p. 120, p. 142].

3 GROUP 3 3 Definition of any hybrid ARQ scheme indicates that retransmission must represent the same set of information bits as the original transmission. However, the set of coded bits can be selected differently as long as transmitted information in each retransmission represents the same set of information bits. Depending on whether the retransmitted bits are required to be the same as the original transmission or not, there exist two main HARQ techniques: Chase combining and Incremental Redundancy [1, p. 120]. In Chase combining (also called HARQ-Type III with one redundancy version [8]) every retransmission contains the same data and parity bits, and receiver uses maximumratio combining to combine the received bits with the same bits from previous retransmissions. Chase combining can be viewed as repetition coding since each transmission is an identical copy of the original transmission. No new redundancy is transmitted, hence no additional coding gain is achieved, and every retransmission can be regarded as adding extra energy to the received transmission through an increased Eb/N0. Eb/N0 is a fundamental quantity in telecommunication systems and is defined as the energy per bit to noise power spectral density ratio. Eb/N0 is a normalized signal to noise ratio (SNR) measure [1, p. 120]. The following figure illustrates HARQ with Chase combining scheme. Fig. 2. HARQ with Chase combining technique With Incremental Redundancy (is called HARQ Type II or HARQ Type III if each retransmission is restricted to be self-decodable [8]), it is not required that each retransmission be exactly identical to the original transmission, and every retransmission contains different information than the previous one. Instead, multiple sets of coded bits are generated, and each set represents the same set of information bits. Typically, whenever retransmission is needed, a set of coded bits that is different from the previous set is used accompanied by different versions generated through puncturing the decoder output. This leads to gaining extra knowledge at every retransmission. Different redundancy versions, i.e. different sets of coded bits, are generated as part of the rate-matching mechanism. The rate matcher uses puncturing (or repetition) to match the number of code bits to the number of physical channel bits available. HSDPA uses Incremental Redundancy HARQ as the basic HARQ technique, and applies puncturing and turbo codes. Puncturing is a process used in coding theory which eliminates some of the parity bits after encoding with an error correction code. In information theory, turbo codes are a class of high performance FEC codes that provide channel capacity. Channel capacity is a theoretical maximum for the code rate at which it is still possible to reliably communicate despite a specific noise level [1, p , p. 142]. The figure below depicts the basic scheme for HARQ Incremental Redundancy. Fig. 3. HARQ with Incremental Redundancy Regarding HSDPA, first, the data block is coded with a punctured 1/3 turbo code. Then, the coded block is usually punctured further during each retransmission in a sense that only a fraction of the coded bits are chosen and sent. Each retransmission uses a different puncturing pattern, therefore different coded bits are sent over the communication channel each time. D. Chase Combining vs. Incremental Redundancy Both Chase combining and Incremental Redundancy are supported by HSDPA standards. Chase combining offers diversity gain and is simple in terms of implementation while Incremental Redundancy generally demonstrates a better performance than Chase combining though at the cost of increased complexity and implementation overhead. It has been shown that redundancy in wireless transmission can significantly enhance throughput and improve performance [1, p. 123, p. 149][4, p. 508]. In [7] it has been shown that in HARQ Type II system, Incremental Redundancy substantially outperforms Chase combining. However, there are situation where Incremental Redundancy cannot be a significantly better choice than Chase combining. It is possible to achieve the highest gains for high channel-coding rates and high modulation orders, but concerning low Modulation and Coding Schemes (MCS), Incremental Redundancy does not yield considerable link-level performance. Moreover, [7] shows that in a system which utilizes link adaptation using Incremental Redundancy, large gains cannot be obtained if link adaptation errors are fairly at a low level. Also, as for fading channels, there are some cases where Incremental Redundancy demonstrates poorer performance than Chase combining. III. SIMULATION AND NUMERICAL ANALYSIS A. Environment Our conducted test case simulates having a nodeb connected to a UE via a HS-DSCH channel, sending packets over the channel measuring the Bit Error Rate (BER) over SNR. The first simulation does not take signal interference as well as the rapid variations in channel condition into account. The modulations available in the Coded Modulation Library (CML) are 4QAM (QPSK) and 16QAM. Unfortunately we had hoped to be able to simulate 32 and 64QAM as well but this was not possible.

4 GROUP 3 4 B. Result Our hypothesis is that skipping the link-adaptation within HSDPA might be beneficial in some specific cases compared to the cases in which there are always overhead brought by HSDPA. The motivation behind this is that in HSDPA, the link adaptation will always cost overhead even when the channel conditions are good whereas in HARQ the overhead in terms of retransmissions will only take place when needed. Our first simulation aims to show the main difference between having the basic error detection and correction scheme brought by HARQ in HSDPA, with having a raw uncoded channel propagating without the benefits of HARQ. We hope to see that for some values over the axis the raw uncoded channel will outperform the channel benefiting from all techniques brought by HSDPA. this, we will try to corroborate our hypothesis in a discussion why this may have worked. C. Discussion Given that our hypothesis is based on some simulation tips from Stefan Parkvall, we will refer back to one of his papers [7], where he brings up a discussion in which environment that this hypothesis might be correct. For instance, as for low MCS, Incremental Redundancy is not capable of providing substantial performance at linklevel. As another case in point, if HARQ with Incremental Redundancy is utilized, it would be impossible to achieve rather high gains when the link is not prone to high rate of errors. Moreover, concerning fading channels Incremental Redundancy cannot always beat Chase combining in terms of performance and throughput. Knowing this, since Incremental redundancy dynamically changes the FEC pattern for each new retransmitted data frames, this implies that FEC scheme plays a huge roll when we are talking about the resultant throughput. So Given this knowledge, our hypothesis is strengthened in the sense that there would be some cases where thoroughly skipping FEC scheme, might result in a better performance of HSDPA. Fig. 4. Simulation of Coded vs. Uncoded channels As we can see from the simulation above we can clearly state that HSDPA outperforms the raw uncoded channel in the long run. Specifically, we can see that this clear improvement in performance comes when having a SNR value around 1 for the coded QPSK channel and around 2 db for the coded 16QAM channel. Although, we can also distinguish from the simulation that for a small window between 1 and 2 db, the uncoded channel actually has a lower BER than the coded HSDPA channels. This gives us an indication that our hypothesis might actually come to be true if we manage to modify the CML library including the benefits of HARQ for the uncoded channels. Something that is not included at the moment. Our simulation scheme for the second simulation was to simulate uncoded channels without link-adaptation to reduce the overhead. To ensure that the data is received properly, the channels in our scenario should utilize HARQ. We hoped that HARQ would reduce the overhead enough to give us a higher data-rate than we achieved with HSDPA in our first simulation, as discussed above. We also hope that HARQ will make it possible to transmit data on channels with a low signal to noise ratio even without the link-adaptation of HSDPA. Non succesful: We were not able to modify CML well enough to run the simulation that we were aiming for. Given IV. CONCLUSION Given that our hypothesis is based on some simulation tips from Stefan Parkvall, we will refer back to one of his papers [7], where he brings up a discussion in which environment that this hypothesis might be correct. For instance, as for low MCS, Incremental Redundancy is not capable of providing substantial performance at linklevel. As another case in point, if HARQ with Incremental Redundancy is utilized, it would be impossible to achieve rather high gains when the link is not prone to high rate of errors. Moreover, concerning fading channels Incremental Redundancy cannot always beat Chase combining in terms of performance and throughput. Knowing this, since Incremental redundancy dynamically changes the FEC pattern for each new retransmitted data frames, this implies that FEC scheme plays a huge roll when we are talking about the resultant throughput. So Given this knowledge, our hypothesis is strengthened in the sense that there would be some cases where thoroughly skipping FEC scheme, might result in a better performance of HSDPA. ACKNOWLEDGMENT We would like to thank Stefan Parkvall for his input and tips regarding our technical analysis.

5 GROUP 3 5 REFERENCES [1] E. Dahlman, S. Parkvall, J. Skld and P. Beming, 3G EVOLUTION: HSPA AND LTE FOR MOBILE BROADBAND. 2nd ed, Academic Press, Oxford, UK, [2] I. Islam, S. Hossain, Comparison of Traffic Performance of QPSK and 16-QAM Modulation Techniques for OFDM System, Journal of Telecommunications and Information Technology, Available: [3] 3GPP TR HSDPA, Overall UTRAN Description; Release 5, Available: [4] A. Goldsmith, Wireless Communications, Cambridge University Press, New York, NY, USA, [5] U. Madhow, Fundamentals of Digital Communication, Cambridge University Press, New York, NY, USA, [6] 3GPP TSG RAN WGI, High speed downlink packet access (HSDPA): Physical layer aspects (Release 5), Tech. Rep v , Mar Available: [7] P. Frenger, S. Parkvall, and E. Dahlman, Performance Comparison of HARQ with Chase Combining and Incremental Redundancy for HSDPA, Vehicular Technology Conference, IEEE VTS 54th, Fall [8] Performance Comparisons of Hybrid-ARQ Schemes, Stockholm, Sweden, Source: Motorola, Available: V. APPENDIX Review Question: Name the two error correction/detection schemes supported by HSDPA? Answer: Chase Combining and Incremental Redundancy