3GPPR1-050386-ViewsonOFDMParameterSet.doc

3GPP TSG RAN WG1 Meeting #41 R1-050386
Athens, Greece, 9 - 13 May, 2005

Source: NTT DoCoMo, Ericsson, Fujitsu, Mitsubishi Electric, NEC, Nortel,
Panasonic, Texas Instruments
Title: Views on OFDM Parameter Set for Evolved UTRA Downlink
Agenda Item: 13.2
Document for: Discussion and Decision
1. Introduction
At the RAN1#40-bis meeting (Beijing), most companies proposed using OFDM-based radio access in the downlink [1]-[17]. This contribution presents our views on the requirements for deciding the OFDM parameter set and the design of OFDM parameter set in the OFDM-based downlink radio access for Evolved UTRA.
2. Requirements for OFDM Parameter Set
2.1. Backward compatibility
Considering the simultaneous use of UTRA and Evolved UTRA (i.e., dual-mode usage) and backward compatibility, the radio-frame length of Evolved UTRA should be identical to that of UTRA, i.e., a 10-msec radio frame.

2.2. Scalable multiple transmission bandwidth
At the RAN Long Term Evolution (LTE) meeting in March 2005 [18], multiple transmission bandwidths from 1.25 MHz to 20 MHz were adopted. The same sub-carrier spacing, i.e., the same useful OFDM symbol duration, is desirable for efficiently supporting a multiple transmission bandwidth. Therefore, the number of sub-carriers is changed according to the transmission bandwidth.

2.3. RAN latency
In the requirements for Evolved UTRA and UTRAN agreed to at the RAN LTE meeting in March 2005 [18], the requirement for the RAN latency (RAN round trip time (RTT)) was decided to be within 10 msec. This RAN latency requirement affects the transmission timing interval (TTI) length. The RAN latency is categorized into the air-interface delay, media delay, and delay in the RNC (or corresponding node above base station (BS), such as the Advanced Access Router (AAR)) as shown in Fig. 1. Furthermore, the air-interface delay in the RAN is classified into the transmitter processing delay, retransmission delay, and receiver processing delay. Here, we assume the delay figures for these three delays as shown in Fig. 2.

Figure 1 – Categorization of RAN Latency

Figure 2 – Air interface delay

Assuming that the transmission delay of an optical fiber is 5 usec/km [19], the media delay, when the distance between the RNC (AAR) and BS is 50 km, becomes 0.25 msec. Moreover, we assume that the delay in the RNC (AAR) is approximately 0.2 msec. Thus, the overall RAN latency can be calculated as

» 2 x (6.5 TTI + 0.5) msec. (1)

Thus, the total RAN latency depends on the TTI length, and the TTI lengths should be designed so that the requirement of the RAN latency of less than 10 msec is satisfied.

2.4. High data rate
At the RAN1#40-bis meeting, many companies proposed the use of a Cyclic Prefix (CP) for OFDM-based downlink transmission [1, 2, 4], although other potential techniques such as offset QAM using the Isotropic Orthogonal Transform Algorithm (IOTA) filter [20] were also proposed. In this paper, we assume the use of the CP in the design of the OFDM parameter set. In the case of OFDM-based radio access, it is clear that the achievable data rate is partly dependent on the CP overhead ratio. This means that, according to the increase in the CP overhead ratio, the achievable data rate is reduced. Therefore, naturally, a small CP overhead loss is necessary to improve the achievable data rate.

2.5. Influence of Doppler effect and phase noise
u Influence of Doppler effect
In the case of OFDM-based radio access, the sub-carrier spacing is designed to be narrower than the channel coherence bandwidth so that the fading of each sub-carrier becomes approximately flat, i.e., frequency-non-selective. Meanwhile, in Evolved UTRA and UTRAN, the maximum user equipment (UE) speed supported should be approximately 350 km/h implying a maximum Doppler spread of approximately 650 Hz and 840 Hz at a 2 GHz and 2.6 GHz carrier frequency, respectively. To mitigate the influence of the Doppler effect, a sub-carrier spacing at least in the range of 10-20 times the maximum Doppler frequency is necessary. In addition to frequency fluctuation due to the Doppler effect, frequency drift due to the frequency difference in the oscillators between the BS and UE occurs. However, we do not consider the influence of frequency drift because the frequency drift becomes small as the reference oscillator in the UE tracks the BS carrier frequency.

u Influence of phase noise
Phase noise is caused by random fluctuations in the frequency of the local-oscillators of the BS and UE. The phase noise has two different kinds of effects: common phase error and inter-sub-carrier interference. Since the influence of the inter-sub-carrier interference is larger than the common phase error and the common phase error can be compensated by means of pilot-aided channel estimation, we only have to take into account the influence of the inter-sub-carrier interference. As a result, sub-carrier spacing should be sufficiently wide so that the influence of inter-sub-carrier interference is small.

2.6. Wide-area coverage support
Wide-area coverage is one of the most important requirements for Evolved UTRA [18]. We roughly categorize support environments as shown in Fig. 3. In an urban area, a large amount of traffic is gathered in a relatively small site-to-site distance area of less than a few kilometers. However, we do not think that the OFDM parameter set should focus on local areas such as hotspots, very-small cells, and indoors with a small cell size. The OFDM parameter set must be optimized considering wide-area coverage support.

Figure 3 – Categorization of environment supported by Evolved UTRA

2.7. High rate data provision of Multicast/Broadcast (MBMS)
As discussed in many contributions at the RAN1#40-bis meeting, OFDM radio access has a beneficial feature in that soft combining of incoming signals from multiple cell sites is easily achieved, improving the received signal-to-interference plus noise power ratio (SINR) particularly at the cell boundary [1, 2, 4, 6]. Using such soft combining, high-data-rate Multicast/Broadcast services can be provided with wide-area coverage. To provide high-data-rate Multicast/Broadcast services, a longer CP duration is necessary for accommodating paths with long time delays from far cell sites and for compensating for inaccuracies in the BS timings.

2.8. Fewer options
A small CP length is desirable for efficient transmission, e.g., to achieve high data rates and high spectral efficiency. On the other hand, a long CP length is necessary for supporting large time dispersion and broadcast with soft combining in large cells. Thus, to achieve an efficient CP overhead ratio according to cell environments and service types (e.g., both Unicast and Multicast/Broadcast), multiple parameters are needed. In other words, it is very difficult to provide a single optimum OFDM parameter set to support very-small to very-wide area coverage up to several tens of kilometers and both Unicast and Multicast/Broadcast services. However, a small number of OFDM parameter sets is desirable for simplifying the implementation and testing equipment. Thus, we present the need for two OFDM parameter sets with different CP lengths.
* Basic short CP length is for typical Unicast environments
* Long CP is for Multicast/Broadcast and for extraordinary large delay spread environments

2.9. PAPR
Since the sub-carrier spacing is designed to be very narrow, the number of sub-carriers becomes several hundred. Then, it is considered that the impact of the difference in the number of sub-carriers from the viewpoint of the peak-to-average power ratio (PAPR) is small assuming such a large number of sub-carriers.

3. Considerations on Basic OFDM Parameter Set for Unicast
3.1. TTI length
From Eq. (1), we find that a TTI length substantially less than 2 msec is necessary to achieve a RAN latency of less than 10 msec. The candidates of the minimum TTI length are then, in our view, 0.5 msec, 0.625 msec, and 0.667 msec. Among these three candidates, the 0.667 msec TTI is identical to the slot length of WCDMA. Furthermore, the number of TTIs per radio frame (= 10 msec) becomes an even number for the 0.5-msec and 0.625-msec TTI lengths. Furthermore, multiple TTI lengths, being multiples of a minimum TTI length, are very beneficial for increasing the payload size especially for a narrow transmission bandwidth such as 1.25 MHz and for reducing the control signaling overhead when a long RAN latency is allowed.

Thus, our conclusions for the radio frame and TTI length are as follows.
- A TTI length of significantly shorter than 2 msec is necessary to achieve a RAN latency of less than 10 msec.
- The radio frame length should be 10 msec for harmonization with the WCDMA parameter set up to Release 6 and the 10-msec radio frame should thus be divisible into multiple TTIs.
- The support for multiple TTI lengths, being multiples of a basic minimum TTI length, is beneficial.
- The recommended candidates for the basic minimum TTI lengths are 0.5 msec, 0.625 msec, and 0.667 msec. Of these, one value should be selected.

3.2. CP overhead ratio
The achievable maximum data rate and throughput depend on the CP overhead ratio. It is desirable to set the CP overhead loss to approximately less than 10% at largest from that without the CP in order to achieve a competitive peak data rate and spectrum efficiency compared to other wireless technologies, although a CP overhead ratio as small as possible is desirable.

3.3. Sub-carrier spacing
We investigated the required sub-carrier spacing considering the influence of the Doppler spread and phase noise.

(1) Influence of phase noise
Phase noise is due to fluctuations in the frequency of the local oscillator(s). Typically, phase-locked-loop (PLL) techniques are applied to frequency synthesizers. Then, the frequency of voltage controlled oscillator (VCO) is locked to a stable reference frequency, which is produced by a temperature-compensated crystal oscillator (TCXO) for a UE terminal (note that there is phase noise in the BS transmitter as well, however, in general the amount is smaller than that in UE terminal). Thus, the jitter of the VCO is suppressed within the tracking loop bandwidth of the PLL. The resultant phase noise is determined by the phase noise of the reference oscillator, divider, and phase detector, etc., which is much smaller than that of a free-running VCO. Figure 4 shows the power density of the phase noise of the VCO using PLL techniques [21]. By using the power density of the phase noise, P(f), the signal-to-interference power ratio (SIR) due to the phase noise as a function of the sub-carrier spacing, Df, is calculated as

(2)

where W(f) is the inter-sub-carrier-interference power as a function of the frequency shift, which is given by

(3)

From Fig. 4 and Eqs. (2) and (3), the SIR due to phase noise is calculated as shown in Fig. 5. The vertical axis on the right-hand side indicates the increase in the required SNR when the received SNR is 15 dB and 22 dB. Figure 5 shows that when the sub-carrier spacing is wider than approximately 10 kHz, the influence of phase noise is small compared to the other impairments.

Figure 4 – Example of power spectrum density of phase noise of VCO using PLL

Figure 5 – SIR due to phase noise as a function of sub-carrier spacing

(2) Influence of Doppler effect
Figures 6(a)-6(b) show the throughput performance as a function of the average received signal energy per symbol-to-noise power spectrum density ratio (Es/N0) with a sub-carrier spacing, Df, as a parameter assuming a fading maximum Doppler frequency of 55.6 Hz corresponding to the moving speed of 30 km/h at a 2-GHz carrier frequency. Figure 6(b) is an enlarged version of Fig. 6(a) focusing on 64QAM modulation. Similarly, Fig 7 indicates the corresponding throughput performance with the fading maximum Doppler frequency of 648.1 Hz corresponding to the moving speed of 350 km/h at a 2-GHz carrier frequency. The Vehicular-A model is used for the figures. Figure 6(b) shows that at low mobility, the sub-carrier spacing should be less than approximately 17 kHz to suppress the loss in the achievable throughput from that with 6-kHz sub-carrier spacing to a low level (e.g., the loss in the throughput from that of 6 kHz is within approximately 1 Mbps (5%)). Furthermore, Fig. 7(b) shows that when the sub-carrier spacing, Df is less than approximately 10 kHz, the required average received Es/N0 at the moving speed of 350 km/h is degraded compared to the case with low mobility especially for 64QAM modulation. More specifically, the sub-carrier spacing should be more than approximately 11 kHz to suppress the loss in the achievable throughput at 350 km/h from that at 30 km/h to within less than approximately 0.5 Mbps (2%). Furthermore, when Df is smaller than 11 kHz, the required Es/N0 loss at 16-Mbps throughput for instance becomes larger than 0.5 dB. Then, we find that a Df value of more than approximately 11-15 kHz, i.e. approximately 20 times the Doppler frequency, is necessary for supporting the moving speed up to approximately 350 km/h, while keeping the loss in the achievable throughput at low mobility to a low level.

(a) 16QAM, R = 2/3 and 64QAM, R = 3/4

(b) Enlarged version of Fig. 6(a) for 64QAM, R = 3/4
Figure 6 – Throughput performance with sub-carrier spacing as a parameter, 30 km/h

(a) 16QAM, R = 2/3 and 64QAM, R = 3/4

(b) Enlarged version of Fig. 7(a) for 64QAM, R = 3/4
Figure 7 – Throughput performance with sub-carrier spacing as a parameter, 350 km/h

3.4. CP length
The root mean square (r.m.s.) delay spread can, according to [22], be well approximated by:

trms = T1 de y (usec) (4)

Table 1 gives the definitions and figures for each of the parameters in Eq. (4).

Table 1 – Definition of parameters in delay spread model
Notation
Definition
Figure
T1
Median value of r.m.s. delay spread at the distance of 1 km (usec)
Urban: 0.4 - 1.0 usec
Suburban: 0.3 usec
Rural: 0.1 usec
Mountain: ³ 0.5 usec
d
Distance between transmitter and receiver (km)
Arbitrary
e
Parameter depending on environment
Urban, Suburban, Rural: 0.5
Mountain: 1.0
y
Random valuable following a log-normal distribution with the standard deviation of sy
sy = 2 - 6 dB

Using the r.m.s. delay spread value assuming the exponentially-decayed power delay profile model as shown in Eq. (5), we calculated the received SINR (here “N” is the background noise) based on the method in [23] as shown in Fig. 8.

p(t) = (1/s) x exp( -t /s )　 s: r.m.s. delay spread (5)

Figure 8 – Useful signal power and inter-symbol and inter-sub-carrier
interference power of received signal.

In Tables 2 and 3, the simulation conditions and the assume