GlobalSIP 2015 -- Symposium on Signal Processing in Mobile Multimedia Communication Systems

Effect of Noise Variance Estimation on Channel Quality Indicator in

LTE Systems
A. M. Mansour, Abd El-Rahman Nada and Ahmed Hesham Mehana

Abstract— The channel quality indicator (CQI) is an impor-

tant LTE parameter that is used to cope with the channel
variations. The accuracy of the CQI calculation depends on
the channel and the noise variance estimations. In this paper,
we study the effect of the noise variance estimation on the
CQI calculation. We assess the performance of a widely-used
CQI algorithm, the mutual information effective SINR mapping Fig. 1: Link Adaptation Model
(MIESM), using two different methods for the noise variance
estimation. The first method is a cyclic-prefix (CP) (time) based
estimation and the second one is a pilot (frequency) based The noise variance estimation, or the signal-to-noise ra-
estimation. It is shown that the CP-based method is more robust
tio (SNR) estimation, affects many system functionalities,
to the channel estimation error compared to the pilot-based one,
and can provide more accurate results especially in the cases e.g. adaptive modulation, turbo coding, channel estima-
where the number of pilots are much less than the CP length tion, equalization such as the minimum mean-square error
(e.g. for small bandwidth). However, the pilot-based method is (MMSE), etc... [2–5]. Several SNR estimation algorithms
much more robust to the presence of interference signals in have been proposed over Rayleigh fading channels [6–9].
the LTE band, a common scenario that is a part of the LTE-
These algorithms can be classified into: data aided (DA)
standard conformance tests.
estimators, which requires pilot symbols as a priori known
transmitted symbols, and non-data aided (NDA) estimators,
which can be blindly applied on unknown transmitted data.
I. I NTRODUCTION Most of the derived techniques were designed for single
input single output (SISO) systems under constant or slowly
time varying channels [10], [11]. This includes the fourth
3GPP Long Term Evolution (LTE) promises high data order and sixth order moment based approaches as presented
rate reliable communication using single and multi-carrier in [12], [13], respectively. There are also many results on
transmissions to combat multipath fading. The multi-carrier estimating the instantaneous SNR over fast time varying
orthogonal frequency division multiplexing (OFDM) based channels in case of SISO, single input multiple output
transmission is used in the downlink. When combined with (SIMO) and multiple input multiple output (MIMO) as in
adaptive modulation and coding (AMC), OFDM allows [14–19].
delivering a reasonable throughput to all users with minimum Using the CQI, the LTE transceiver can reliably adapt with
requirements on Quality of service (QoS) [1]. Link adapta- the channel variations to optimize the system performance
tion (LA) is often used to cope with the channel variations by selecting the appropriate code rate and modulation order
as shown in Fig.1. LA adapts the communication rate to specified by the CQI index. This index is calculated by the
the channel variations by selecting the suitable modulation UE such that the block error rate does not exceed 10% [20].
and coding scheme (MCS) using the feedback information In this paper, we investigate the effect of the noise variance
obtained from the user equipment (UE). The channel quality estimation on the CQI selection and most importantly the
indicator (CQI) is an important LTE metric that allows system performance in terms of the achievable throughput.
successful LA. The UE performs some measurements, e.g. We present two algorithms for the noise variance estimation.
channel and noise variance estimations, based on which the The first algorithm is performed in the frequency domain
CQI is calculated and fed back to the base station. In order using a known pilot sequence, whereas the second one
for the UE to provide a reliable CQI estimate, it has to have is performed in the time domain using the cyclic-prefix
reliable channel and noise variance estimates. available in the received signal.

II. S YSTEM M ODEL

A. Mansour is with Department of Electronics & Communication Engi- The LTE system uses OFDM with bandwidth that varies
neering, and Center of Smart Nanotechnology & Photonics (CSNP), Smart
CI Research Center of Excellence, Alexandria University, Egypt, Email: from 1.4 MHz (corresponding to 72 subcarriers) up to 20
ahmed mansour@mena.vt.edu MHz (corresponding to 1200 subcarriers) according to the
A. Nada and A. Mehana are with Department of Electronics, Electrical LTE standard specification [20]. The UE should be able to
Communication Engineering Department, Cairo University, and Wireless
Advanced System Innovations and Electronics Art (wasiela), Egypt, Email: make the CQI measurements within the transmission time
abdulrahman.nada@wasiela.com, and Email: ahesham@ieee.org interval (TTI) over the utilized bandwidth for a number

978-1-4799-7591-4/15/$31.00 ©2015 IEEE 156

 GlobalSIP 2015 -- Symposium on Signal Processing in Mobile Multimedia Communication Systems

of resource blocks (RB). A resource block consists of 12 We note here that, the CP-based noise variance estimation
subcarriers and (6 or 7) OFDM symbols in case of normal provides a single estimate for the time-frequency grid while
CP or extended CP respectively [21]. Fig.2, presents a sim- the pilot-based algorithm can provide multiple noise variance
plified model for an LTE downlink system. The transmitter estimates depending on the portion of the time-frequency
chain usually consists of segmentation and cyclic-redundancy grid over which the estimation takes place. Hence the pilot-
check (CRC) attaching followed by channel coding and based algorithm is more suitable for the subband CQI cal-
modulation that assigns a pair of code-rate and modulation culation where multiple noise variance estimates are needed,
order (e.g. 4/16/64-QAM modulation) that corresponds to the i.e. one per subband. This is especially important in the case
CQI feedback from the UE. The modulated symbols are then of the existence of an interference over the time-frequency
mapped onto the time-frequency grid . The final stage of the grid. We provide more details on this scenario in the sequel.
transmitter is the IFFT stage and a cyclic-prefix insertion. Denoting the OFDM symbol number by m, the received
sampled signal is given by:

L−1
ym (n) = hm (l)xm (n − l) + wm (n) (1)
l=0

where L is the number of fading channel paths, hm (l) is

the -th channel coefficient in the m-th OFDM symbol,
wm (n) ∼ CN (0, σ 2 ) is the n-th sample of the additive white
Gaussian noise (AWGN), and σ 2 is the noise variance. The
Fig. 2: Simplified model of LTE System k-th sample of the transmitted signal in the m-th OFDM
symbol is denoted by xm (k). The following sections present
the noise variance estimation based on: the cyclic prefix (CP)
Fig.3, presents the grid of resource elements (RE) distri- in the time-domain, as well as the pilot sequence in the
bution in the case of normal CP of two antenna ports. It frequency-domain.
shows allowable positions of reference signal (RS) of each
antenna port, which means the allowable position of antenna III. C YCLIC -P REFIX BASED N OISE VARIANCE
port one cannot be used for antenna port two, and vice versa E STIMATION
[21].
For wireless communication systems, the delay spread
caused by multipath fading channel introduces ISI to the
receiver. OFDM transmission mitigates the multipath effects
by transforming the frequency-selective channel into a paral-
lel set of flat fading sub-channels. The cyclic-prefix is used
to eliminate both inter symbol interference (ISI) and inter
carrier interference (ICI) providing that the CP length is
greater than the channel delay spread. Since the CP contains
a repeated version of a part of the OFDM symbol, it can
be used for the noise variance estimation [2]. Estimating
the noise variance is performed by multiplying the received
signal (cf. Fig. 4.)
Fig. 3: Mapping of the downlink reference signals with
M Ng −1
normal CP and two antenna ports |ym (n) − ym (n + N − Ng )|2
 = m=1 n=dmax +1
σ 2
2M (Ng − dmax )
The receiver chain consists of an FFT stage and cyclic (2)
prefix removal followed by channel and noise variance where M is the number of OFDM symbols used in the
estimations. The estimated channel and noise variance are calculation, 1 N is the number of samples in the OFDM
used for the CQI calculation, as well as the data detection symbol, Ng is the number of the CP samples, and dmax is
and decoding. The UE sends the CQI measurements to the the maximum delay spread. The reason we ignore the first
transmitter over physical uplink control channel (PUCCH) dmax samples of the CP is that the estimates from these
that indicates an appropriate modulation and coding scheme samples are not reliable due to the ISI effect.
(MCS) corresponding to the current channel state as shown
in Fig.1.
The standard specifications [20] provide two types of CQI
calculation: wideband and subband. Wideband CQI calcula- Fig. 4: CP Inserstion
tion provides one CQI for the whole band, whereas subband
CQI calculation provides multiple CQIs, i.e. one CQI per 1 The number of OFDM symbols in an LTE-FDD subframe is either 12
each subband. or 14 in the case of normal or extended CP respectively

157
 GlobalSIP 2015 -- Symposium on Signal Processing in Mobile Multimedia Communication Systems

The variable dmax has to be calculated accurately since based on which the normalized mutual information per
too large dmax decreases the ISI effect but also decreases the received bit (RBIR) is evaluated
estimation reliability (by Cramer-Rao bound), whereas too 
N
small dmax introduces more ISI in the case of frequency- SI(SIN Rn , m(n))
n=1
selective channel. We assess the selection of dmax by means RBIR = . (5)
of simulations in Section VI. 
N
m(n)
n=1
IV. P ILOT-BASED N OISE VARIANCE E STIMATION The SN Ref f is obtained from RBIR using numerically-
Although the CP-based noise variance estimation is sim- calculated tables. One of the advantages of the MIESM
ple, it is not robust to in-band interference. In the following algorithm is that the curves/tables of the SI(SIN Rn , m(n))
we present a frequency-domain based method for the noise and the RBIR are generated once in the system simulator
variance estimation. Since the pilot sequence (values and for each modulation order and sorted in a look-up table for a
positions in the time-frequency grid) is known to the user given SNR range. In LTE, there are 16 different CQI values
equipment, the noise variance is estimated as (from 0 to 15) each indicating a modulation order and a
coding rate pair. [20].
1 
2 =
σ  n,m Xn,m
|Yn,m − H ∗
|2 (3) VI. S IMULATION R ESULTS
Ntotal n m
This section presents Monte Carlo simulations for the
where Yn,m is the received signal after the FFT operation. noise variance estimation in LTE system and its effect on
The indices n and m denote the pilot index and the OFDM the CQI calculation. We provide three different performance
symbol index respectively. H  n,m and Xn,m are the channel measures. The first one is the effect of the shift value, dmax
estimate and the pilot symbol over the (n, m) frequency- in the CP-based estimation. The second one is the normal-
time grid, respectively. Ntotal is the total number of available ized mean square error of both CP-based and pilot-based
pilots used in the estimation process. Eq. (3) can be easily estimations. The third one is the LTE achieved throughput
extended for multiple transmit and/or receive antennas. in case of CQI reporting for AWGN channel.
We note that the noise variance estimation can be per- A. The CP-Based Shift Selection
formed on separate bands of the occupied bandwidth, making
In this test, we compare between the CQI obtained using
this algorithm more suitable in the case of colored interfer-
the CP-based noise variance estimation and that obtained
ence. Obviously, though, dividing the occupied bandwidth
when the noise variance is assumed perfectly known. The
into many bands has the accuracy-robustness trade-off. Even
(difference) error between these two CQI indices is as-
in the presence of white interference, i.e. the interference oc-
sessed in three LTE defined channels, ETU-300Hz, EPA-
cupies the same bandwidth as the desired signal, this method
5Hz and EVA-70Hz [30], [31]. Three different shift values
is more suitable for noise plus interference estimation, es-
pecially in synchronous transmissions where multiple base- are tested:0, 20, and 85. These values are chosen based on
stations in multiple cells transmit their signals synchronously extensive simulations that we omit in this paper. Fig.5 shows
to their users using the same time-frequency resources [22], that the error varies from -1 to 1 for the selected shift values.
[23]. It is also shown that the shift value 85 is suitable for all three
channels and gives the least error.
V. CQI C ALCULATION : M UTUAL I NFORMATION
E FFECTIVE SINR M APPING M ETHOD
This section presents the most common technique used
for the CQI calculation, the mutual information effective
SINR mapping (MIESM) method [24–26]. Basically, the
SIN R at all subcarriers of the time-frequency grid (Fig.3)
are mapped into a single (or multiple) values referred to as
the effective SNR, SN Ref f , which is then used to find an
estimate of the CQI from basic SISO AWGN block error
rate (BLER) curves [27–29]. This is done as follows. The Fig. 5: The (difference) error between the CQI calculated
2
n-th subcarrier SIN Rn is calculated asSIN Rn = |Ĥ| σ̂ 2 .
using the noise variance estimation and that calculated with
Then, the symbol mutual information (SI) is calculated a perfect knowledge of the noise variance. Three LTE-defined
  channels are used with three different shift.
1
M
SI(SIN Rn , m(n)) = log2 (M ) − M EU log2 1+
m=1 B. Normalized Mean Square Error

M   Fig.6, shows the normalized mean square error of the
|Xk − Xm + U |2 − |U |2
exp − (4) noise variance estimation using the CP-based and the pilot-
(1/SIN Rn ) based methods in 10 MHz LTE bandwidth over ETU-0Hz.
k=1,k==m

158
 GlobalSIP 2015 -- Symposium on Signal Processing in Mobile Multimedia Communication Systems

We simulate different cases for the antenna setup nt × nr show a saturation in the throughout curves in the case of no
where nt and nr are the numbers of transmit and receive CQI reporting as expected.
antennas, respectively. The noise signal is assumed additive 55

white Gaussian. It is shown that the CP-based method can 50

No CQI Feed
back with MCS = 6
CQI Feed
back with Pilot
based

provide more accurate results compared to the pilot-based 45 CQI Feed
back with CP
based

40

one in the cases where the SNR is low and the number of 35

Throughput (Mbps)
pilots are less than the CP length, e.g. for SISO and low SNR. 30

25
By using more antennas, the number of observations (pilots) 20

increases and the performance of the pilot-based method 15

approaches (e.g. SIMO at low SNR) or even exceed that 10

5
of the CP-based method (e.g. MIMO or in the high SNR). 0

5
3
1 1 3 5 7

In this simulation we assume that the channel is perfectly SNR (dB)

known and the CP size is normal. Fig. 8: The throughputs in an AWGN channel in the cases
In Fig.7, we simulate the case of 10 MHz LTE band- of: no feed-back, CQI with pilot-based and with CP-based
width over EPA-5Hz with/without perfect channel knowledge methods.
and in the presence of a synchronous interfering 10 MHz
downlink (DL) LTE signal. In this case, the performance of 40

the pilot-based is superior even in the presence of channel 35

No CQI Feed
back with MCS = 6

estimation error. The reason is detailed in the next section. 30

CQI Feed
back with CP
based
CQI Feed
back with Pilot
based

3 25
10

Throughput (Mbps)
20
2 1X1 Pilot based
10
1X2 Pilot based 15
2X2 Pilot based
4X2 Pilot based
1 10
10 CP based

5
Normalized MSE

0
10
0

−1
5
4
3
2
1 0 1 2 3 4
10 SNIR (dB)

−2
10 Fig. 9: The throughputs in an AWGN channel and an LTE
−3
10
Down-link synchronous interference in the case of: no feed-
back, CQI with pilot-based and with CP-based methods.
−4
10
−5 0 5 10 15 20 25 30
SNR (dB)

When the CQI algorithm is engaged, the throughput

Fig. 6: NMSE of both estimation methods with perfect
increases with SNR in an identical manner for both CP-
channel estimation, white noise, and ETU-0Hz channel
based and pilot-based algorithms for white Gaussian noise.
However, when a synchronous LTE interference signal exists,
the CP-based performance is catastrophic as shown in Fig. 9.
The reason for such performance is as follows. The CP-
−1
10 based algorithm cannot detect the interference power in
synchronous scenario because the LTE interfering signal has
Normalized MSE

the same structure as that of the LTE desired signal. Thus

1X1 Pilot based − Channel Estimation Algorithm
when the CP-based algorithm subtracts the beginning and
−2
10 1X2 Pilot based − Channel Estimation Algorithm
2X2 Pilot based − Channel Estimation Algorithm
the end of the received signal, only the noise signal remains
4X2 Pilot based − Channel Estimation Algorithm
CP based and its power is estimated [32], [33]. Using this misleading
1X1 Pilot based − Perfect Channel Estimation
1X2 Pilot based − Perfect Channel Estimation
2X2 Pilot based − Perfect Channel Estimation
noise estimate as the noise plus interference estimate results
−3
10
4X2 Pilot based − Perfect Channel Estimation
in an optimistic (high) CQI that does not reflect the actual
−5 0 5 10 15 20 25 30
SNIR (dB) interference situation. The corresponding MCS results in
Fig. 7: NMSE of both estimation methods, with white noise erroneous transmission and thus low throughput. We omit
and LTE synchronous interference, and EPA-5Hz channel the results for the subband CQI due to the space limitations.
VII. C ONCLUSION
C. Total Achieved Throughput In this paper, we study the effect of the noise variance
This section presents a comparison between the achieved estimation on the CQI calculation in LTE systems using
throughputs for the following cases: CP-based CQI, pilot- the mutual information effective SINR mapping (MIESM)
based CQI, and no CQI feedback (i.e. there is no adaptive algorithm. We use the cyclic-prefix (CP) (time) based and
MCS). The 1 × 2 SIMO system uses 10 MHz bandwidth, the pilot (frequency) based noise variance estimation. While
over AWGN. Fig. 8 and Fig.9 show the throughputs for white the first one is more robust to the channel estimation errors,
Gaussain noise only, and white Gaussian noise plus 10 MHz the second one is more robust to the presence of interference
LTE-downlink interference signal, respectively. These figures signals but is limited to the number of pilots.

159
 GlobalSIP 2015 -- Symposium on Signal Processing in Mobile Multimedia Communication Systems

R EFERENCES [26] Ericsson, S. Tsai and A. Soong, Effective-SNR mapping for modeling
frame error rates in multiple-state channels, 3GPP2-C30-20030429-
[1] Xiaowen Li, Qianjun Fang, Liuwei Shi,”A Effective SINR Link to 010, Ericsson. Standardization Document, 3rd generation partnership
System Mapping Method for CQI Feedback in TD-LTE System,” project2 3GPP2, 2003.
Chongqing Key Lab. of Mobile Commun., Chongqing Univ. of Posts [27] Xiaowen Li, Qianjun Fang, Liuwei Shi,”A Effective SINR Link
& Telecommun. (CQUPT), Chongqing, China - 2011 to System Mapping Method for CQI Feedback in TD-LTE Sys-
[2] Tao Cui and Chintha Tellambura,”Power Delay Profile and Noise tem,”Chongqing Key Lab. of Mobile Commun., Chongqing Univ. of
Variance Estimation for OFDM,” IEEE Commun. Lett., VOL. 10, NO. Posts & Telecommun. (CQUPT), Chongqing, China - 2011
1, Jan. 2006. [28] Wang Xiaoqin and Wang Xin, ”Efficient Link Quality Prediction for
[3] N. C. Beaulieu, A. S. Toms, and D. R. Pauluzzi, Comparison of four OFDM Systems,” IEEE Vehicular Technology Conference (VTC),
SNR estimators for QPSK modulations, IEEE Commun. Lett., vol. 4, 2012.
pp. 43-45, Feb. 2000. [29] IEEE 802.16m Evaluation Methodology Document (EMD)- IEEE
[4] N. S. Alagha, Cramer-Rao bounds of SNR estimates for BPSK and C802.16m-08/119
QPSK modulated signals, IEEE Commun. Lett., vol. 5, pp. 10-12, Jan. [30] 3GPP TS 36.101. ”User Equipment (UE) Radio Transmission and Re-
2001. ception.” 3rd Generation Partnership Project; Technical Specification
[5] P. Gao and C. Tepedelenlioglu, SNR estimation for nonconstant Group Radio Access Network; Evolved Universal Terrestrial Radio
modulus constellations, IEEE Trans. Signal Process., vol. 53, pp. 865- Access (E-UTRA).
870, Mar.2005. [31] 3GPP TS 36.104. ”Base Station (BS) radio transmission and re-
[6] S. He and M. Torkelson, Effective SNR estimation in OFDM system ception.” 3rd Generation Partnership Project; Technical Specification
simulation, in Proc. GLOBECOM, Nov. 1998, pp. 945-950. Group Radio Access Network; Evolved Universal Terrestrial Radio
[7] X. Xu, Y. Jing, and X. Yu, Subspace-based noise variance and SNR Access (E-UTRA).
estimation for OFDM systems, in Proc. WCNC, Mar. 2005, pp. 23-26. [32] Lpez-Prez, David, et al. ”Power minimization based resource alloca-
[8] B. Yang, K. B. Letaief, R. S. Cheng, and Z. Cao, Channel estimation tion for interference mitigation in OFDMA femtocell networks.” IEEE
for OFDM transmission in multipath fading channels based on para- Journal on Selected Areas in Communications, 2014 pp. 333-344.
metric channel modeling, IEEE Trans. Commun., vol. 49, pp. 467-479, [33] Lpez-Prez, David, et al. ”On distributed and coordinated resource al-
Mar. 2001. location for interference mitigation in self-organizing LTE networks.”
[9] S. Boumard, Novel noise variance and SNR estimation algorithm for IEEE/ACM Trans. on Networking (TON), 2013, pp. 1145-1158.
wireless MIMO OFDM systems, in Proc. GLOBECOM, Dec. 2003,
pp.1330-1334.
[10] D. R. Pauluzzi and N. C. Beaulieu, A comparison of SNR estimation
techniques for the AWGN channel, IEEE Trans. Commun., vol. 48,
no. 2, pp. 1681-1691, Oct. 2000.
[11] M. Turkboylari and G. L. Stuber, An efficient algorithm for estimating
the signal-to-interference ratio in TDMA cellular systems, IEEE Trans.
Commun., vol. 46, no. 6, pp. 728-731, June 1998.
[12] P. Gao and C. Tepedelenlioglu, SNR estimation for non-constant
modulus constellations, IEEE Trans. Signal Process., vol. 53, no. 3,
pp. 865-870, Mar 2005.
[13] R. Lopez-Valcarce and C. Mosquera, Sixth-order statisticsbased non-
data-aided SNR estimation, IEEE Commun. Letters, vol. 11, no. 4,
pp. 351-353, Apr. 2007.
[14] A. Das (Nandan), NDA SNR estimation: CRLBs and EM based
estimators, in Proc. IEEE TENCON08, pp. 1-6, 2008.
[15] A. Stephenne, F. Bellili, and S. Affes, Moment-based SNR estimation
over linearly-modulated wireless SIMO channels, IEEE Trans. Wire-
less Commun., vol. 9, no. 2, pp. 714-722, Feb. 2010.
[16] A. Stephenne, F. Bellili, and S. Affes, Moment-based SNR estimation
for SIMO wireless communication systems using arbitrary QAM, in
Proc. 41st Asilomar Conference on Signals, Systems and computers,
Pacific Grove, CA, USA, 2007, pp. 601-605.
[17] M. A. Boujelben, F. Bellili, S. Affes, and A. Stephenne, SNR esti-
mation over SIMO channels from linearly modulated signals, IEEE
Trans. Signal Process., vol. 58, no. 12, pp. 6017-6028 , Dec. 2010.
[18] M. A. Boujelben, F. Bellili, S. Affes, and A. Stephenne, EM algorithm
for non-data-aided SNR estimation of linearly-modulated signals over
SIMO channels, in Proc. of IEEE GLOBECOM, Honolulu, Hawaii,
USA, 2009, pp. 1-6.
[19] A. Das and B. D. Rao, SNR and noise variance estimation for MIMO
systems, IEEE Trans. Signal Process., vol. 60, no. 8, pp. 3929-3941,
Aug. 2012.
[20] Evolved Universal Terrestrial Radio Access (E-UTRA); Physical layer
procedures (3GPP TS 36.213 version 8.8.0 Release 8).
[21] Evolved Universal Terrestrial Radio Access (E-UTRA): Physical chan-
nels and modulation (3GPP TS 36.211 version 8.8.0 Release 8).
[22] Zhang, Zhongshan, et al. ”On swarm intelligence inspired self-
organized networking: its bionic mechanisms, designing principles and
optimization approaches.” IEEE Communications Surveys & Tutorials,
2014, pp. 513-537.
[23] Wang, Chenwei, et al. ”Design and operation of blind interference
alignment in cellular and cluster-based systems.” IEEE Information
Theory and Applications Workshop (ITA), 2011.
[24] Ericsson, System-level evtiluation of OFDM further considerations,
3GPP TSG-RAN-135, Document R1-031303. Lisbon, Portugal, Nov.
2003.
[25] 3GPP2-C30-20030429-010, Effective-SNR Mapping for Modeling
Frame Error Rates in Multiple-state Channels, Ericsson.

160