See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/220731838

Flicker Noise Mitigation in Direct-Conversion Receivers for OFDM Systems

Conference Paper in Acoustics, Speech, and Signal Processing, 1988. ICASSP-88., 1988 International Conference on · April 2009
DOI: 10.1109/ICASSP.2009.4960153 · Source: DBLP

CITATIONS READS
2 141

3 authors, including:

Masoud Sajadieh Hossein Alavi

Intel Intel, Hillsboro, OR, USA
13 PUBLICATIONS 304 CITATIONS 18 PUBLICATIONS 370 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

millimeter waves View project

All content following this page was uploaded by Masoud Sajadieh on 05 March 2015.

The user has requested enhancement of the downloaded file.

 Flicker Noise Mitigation in Direct-Conversion
Receivers for OFDM Systems
Surendra Boppana∗ , Masoud Sajadieh† and Hossein Alavi†
([email protected], [email protected], [email protected])

Abstract— Direct-conversion receiver (DCR) architecture has y = x + nw + nf Output = x + nw + e

received considerable attention recently owing to its portable

architecture and superior performance in terms of power and
cost over its super-heterodyne counterpart. Flicker noise is one
of the major impairments which severely affects the performance u = nw + nf
Prediction Filter
hp Prediction = fp
of the system. In this work, we investigate the use of signal
processing techniques in the mitigation of flicker noise in OFDM-
based systems which employ DCR architecture. The statistical Fig. 1. Flicker noise mitigation through prediction filtering.
properties of flicker noise are exploited to develop adaptive signal
processing algorithms that reduce the effect of flicker noise in
DCRs. Results indicate that signal processing algorithms can approach, we focus on key ideas and issues that impact the
provide significant performance gain under low SNR conditions. performance of the system.
I. I NTRODUCTION II. F LICKER N OISE M ITIGATION IN OFDM- BASED
A major source of degradation in DCRs is flicker noise. SYSTEMS
Flicker noise is present in all electronic devices, and is In this section, we investigate the problem of flicker noise
produced due to a variety of physical phenomenon [1], [2]. mitigation in DCRs in OFDM-based systems. We assume
Flicker noise has a power spectral density which is inversely that flicker noise is the dominant source of impairment in
proportional to frequency. Due to the nature of flicker noise, it the system, and other non-linearities such as dc offset, I/Q
is dominant at frequencies close to zero, and hence impacts the mismatch etc., have already been mitigated. For the sake of
performance of DCRs. The coupling of the flicker noise with exposition, we first consider the case of optimal prediction of
the received signal occurs after down-conversion to baseband. flicker noise assuming that the prediction filter has access to
Since the root-mean squared (rms) power of the received the “noise only” samples. This approach helps us in establish-
signal is in the order of micro-volts, flicker noise comprises ing bounds on the performance of practical filtering schemes,
a substantial fraction of the signal power, which leads to and in identifying the parameters that effect the performance
large signal distortions. The effect of flicker noise can be of flicker noise mitigation algorithms. Later on, we discuss
reduced by at the device level by techniques such as correlated the practical issues involved in implementing the developed
double sampling [3], edge-extended design [4], employing algorithms in DCRs designed for supporting OFDM-based
large gate area devices [2], [5], etc. However, its effect cannot systems. The modeling of flicker noise in this work is based
be completely mitigated. Its effect is more pronounced in on the results reported in [9], [10].
CMOS devices [6].
Flicker noise, also known as pink noise or 1/f noise is A. Optimal Prediction Filtering
intrinsically present in all electronic devices. It is characterized The setup of the optimal prediction filtering scheme is
by a power spectral density (PSD), Sf (f ), which is inversely shown in Fig. 1. The input signal y(k) consists of the desired
proportional to the frequency f , i.e. Sf (f ) ∝ f1 , f > 0. signal x(k), which is assumed to be Guassian, white noise
There has been recent interest in the use of signal processing nw (k), and flicker noise nf (k) all of which are pairwise
techniques in mitigation of non-linearities in DCRs. This is independent. We also assume that the prediction filter hp
motivated by the strong push towards flexible and software has access to the present and all the past noise samples
configurable receiver architectures as more and more func- u[k] = nw [k] + nf [k]. Under this set-up, the optimal filter
tionality of a DCR is performed in the baseband using digital coefficients hp [n], 0 ≤ n < L that minimize the mean square
signal processors. Adaptive signal processing algorithms have error between the noise samples u and the predicted signal fp
been proposed [8] to tackle the non-linearities in DCRs. In are given by the Wiener-Hopf equations [11].
this work, we investigate the use of adaptive signal processing The PSDs of the input signal u(k), and the error signal
algorithms in mitigating the effect of flicker noise in DCRs. e(k) = u(k) − fp (k) are shown in Figs. 2(a) and 2(b),
Although the algorithms considered in this work can be respectively. The PSD of the flicker noise, and the white noise
applied to any DCR architecture, we specifically address the components of the input signal u(k), are shown in Fig. 2(a).
DCR architectures that support OFDM systems such as WiFi The corner frequency fc , of the flicker noise is 1 MHz, the
and WiMAX. Instead of taking a mathematically rigorous bandwidth of the system is 20 MHz, and the length of the

978-1-4244-2354-5/09/$25.00 ©2009 IEEE 2593 ICASSP 2009

 

 
'





 
'







'









'




 

'





 $
 

  
  
!"#

%& 
 


 "



 



 





            
 $ % #" !%$
!"%

Fig. 3. SNR gain due to prediction filtering of flicker noise.



   




(a) PSD of flicker and white noise signals at input of prediction

filter 






 

$
" 
!"%
"#'











!#

! 
"
! 







 







        


     

 
#%
'

(b) PSD of residual flicker and white noise signals

Fig. 4. SNR gain due to prediction filtering of data and noise signal.
Fig. 2. PSD of the flicker and white noise signal before and after predictive
cancellation, fc = 1 MHz.
σ 2 +σ 2
received signal y(k), i,e., SNR gain = σw2 +σf2 , where σe2 is
w e
the variance of the residual flicker noise. Since the input to
prediction filter is 50. We can observe in Fig. 2(a), the PSD the prediction filter is “noise only” samples, the SNR gain due
curve of the flicker noise intersecting the PSD curve of the to prediction filtering is independent of the input SNR, and is
white noise at about 1 MHz. The flicker-to-thermal noise ratio dependent only on the corner frequency of the flicker noise
is -2.7 dB. (and FNR). From Fig. 3, we note that the SNR gain increases
Figure 2(b) depicts the PSD of the residual flicker noise with an increase in the corner frequency. It can be seen from
and the white noise components after the subtraction of the Fig. 3 that for any given corner frequency fc , there is minimal
predicted flicker noise sample fp (k) from the input signal improvement in performance for filter lengths greater than 50.
u(k) (The signal component, x(k) is not shown in this figure).
There are a few of interesting observations to be made. We B. Modified Prediction Filtering
note that the variance of the residual flicker noise is less In this setup, the received signal y(k) acts as the input
than the variance of the input flicker noise component. This to the prediction filter. The prediction filter generates an
is not surprising, since the prediction filter exploits the high estimate fp (k), of the present sample of the flicker noise,
correlation among the flicker noise samples to predict the and subtracts it from the received signal, y(k). This design
current flicker noise value, and cancel it out from the received is motivated by the fact that under low SNR conditions,
signal. It is interesting to note that the PSD of the residual the received signal y(k) is dominated by the flicker noise
noise also follows the inverse power-law, where the PSD is component nf (k), and y(k) can provide a reasonable estimate
proportional to 1/f α , α < 1. of “noise only” samples. Since the input to the prediction filter
The SNR gain due to mitigation of flicker noise through consists of signal and noise components, the performance of
prediction filtering is plotted in Fig. 3 for several values of the prediction filter strongly depends on the input SNR.
flicker noise corner frequencies, fc . The input SNR is 10 dB, The SNR gain due to modified prediction filtering is shown
and is constant over all the corner frequencies simulated. in Fig. 4 for several values of input SNR, and flicker noise
The desired corner frequencies are simulated by appropriately corner frequency fc . For any given corner frequency, the SNR
scaling the variances of the flicker noise and white noise gain due to prediction filtering decreases as the input SNR
components. The SNR gain is defined as the ratio of the increases. As the input SNR increases, the data component
SNRs of the output signal yp (k) = y(k) − fp (k) and the x(k) increases, and affects the performance of the prediction

2594
 6
14
Fullband, SNR = 5 dB
Fullband, SNR = 0 dB
12 5 Subband, SNR = 0 dB
Subband, SNR = 5 dB

10
4

SNR gain (dB)

SNR gain (dB)

8
= 0.25
3

= 0.5
6
= 1.0

= 1.5
2

= 2.0
4

1
2

0
0 0 0.5 1 1.5 2 2.5 3 3.5 4.0
0 0.5 1 1.5 2 2.5 3 3.5 4 Corner Frequency (MHz)
Corner Frequency (MHz)

Fig. 7. SNR gain due to subband prediction filtering of data and noise signal.
Fig. 5. SNR gain due to prediction filtering of flicker noise.

filter (In fact, the desired signal x(k) behaves as white noise prediction filtering is shown in Fig. 6. Analysis/Synthesis [12]
during the prediction of flicker noise). However at low input filters are employed to decompose the input signal y(k) =
SNRs, there is considerable SNR gain achieved. For instance, x(k) + nw (k) + nf (k) into multiple bands, and prediction
when the input SNR is 3 dB, and the corner frequency of the filtering is performed on the band in which flicker noise is the
flicker noise is 2 MHz, we observe an SNR gain of 1.35 dB. most dominant noise source (this corresponds to the band of
Figure 5 depicts the SNR gain achieved due to modified frequencies closest to dc ). The number of analysis/synthesis
prediction filtering of flicker noise for several values of the sections is determined based on the input SNR, flicker noise
power-law exponent α. The input SNR is held constant at corner frequency fc , FNR, and the implementation complexity
3 dB, the corner frequency fc of the flicker noise is 2 MHz, involved. The theory of analysis/synthesis filters is well devel-
and FNR is 0 dB. We note that for a given value of corner oped, and there exits vast literature that deals with various
frequency, SNR gain increases with an increase in α. Increase aspects of the analysis/synthesis filters. In this work, we
in α translates to increased correlation between the present employ Quadrature Mirror Filter (QMF) banks [13] that result
and past samples, and an increase in correlation translates to in perfect reconstruction of the input signal in the absence
better prediction of the current flicker noise sample. of the prediction filtering scheme. There exist a variety of
techniques for the design of the QMF, and in this work we
H1 2 2 G1
employ the design methodology described in [13].
y = x + nw + nf Output = x + nw + es
H2 2 2 G2 The SNR gain due to sub-band prediction filtering of flicker
noise is shown in Fig. 7, for several values of input SNR, and
Prediction Filter
hp fp corner frequency fc . The bandwidth of the system is 20 MHz.
The number of analysis/synthesis sections were chosen such
Fig. 6. Flicker noise mitigation through subband prediction filtering. that the lowest subband contained all the frequencies from dc
until fc . For example, when the corner frequency fc =1 MHz,
we employ four analysis/synthesis filters on the received
C. Sub-band Prediction Filtering signal, and the prediction filtering is applied on the signal
In the previous sections, we have considered the case of in band from dc until 1.25 MHz. Wiener filter of length 50 is
prediction filtering on the entire bandwidth of the system. employed to perform prediction filtering in the subband where
However, due to the characteristics of the flicker noise, it flicker noise is the dominant source of noise. We observe that
is dominant only at frequencies close to dc, specifically, the SNR gain (measured over the entire signal bandwidth) is
until the corner frequency, beyond which white noise is the much greater in the case of subband prediction filtering than in
dominant source of degradation. Based on this observation, the case of fullband prediction filtering. When fc = 1 MHz, and
we investigate prediction filtering schemes that operate in the input SNR is 5 dB, the SNR gain due to subband prediction
band in which flicker noise is dominant, i.e. subband prediction filtering is 1.5 dB compared to 0.8 dB of SNR gain in the
filtering schemes. The prediction filtering mechanism of the case of fullband prediction filtering. The SNR gain is high
subband filtering is quite similar to that of the modified when the input SNR is low, and the SNR gain decreases as
prediction filtering of Section II-B, expect that prediction the input SNR increases. This observation is similar to the
filtering is performed on a low-pass filtered signal in which behavior observed in the case of fullband prediction filtering
flicker noise is the dominant component. The setup of subband described in Section II-B.

2595
 

The SNR gain due to subband Wiener filtering of flicker
noise is shown in Fig. II-E for two different values of flicker

noise. The input SNR of the signal (measured over the entire

band) is 0 dB. When fc = 2 Mhz, we observe an SNR gain of
4.2 dB for subcarriers -1 and 1, and the SNR gain decreases as

 

 the sub-carrier index increases. The SNR gain achieved due

to subband Wiener filtering is slightly greater than the gain


 achieved due to subband prediction filtering (refer to Fig. 8).



 

III. C ONCLUSION
In this work, we investigated the use of adaptive filtering


      
schemes in mitigating flicker noise in DCRs supporting OFDM


systems. The strong correlation among the flicker noise sam-
ples can be exploited to perform prediction filtering of flicker
Fig. 8. Sub-band Prediction filtering of flicker noise, fc =2 MHz noise, and the predicted signal subtracted from the received


signal. However, we observed that this approach works only
when the input SNR is low. Subband prediction filtering can be
employed to improve the performance of prediction filtering.
 Substantial gains can be achieved when the flicker noise

corner frequency is high. We also considered flicker noise
mitigation through Wiener filtering of the received data in

OFDM systems. We observed that subband Wiener filtering

!"

 is superior to subband prediction filtering and can provide


significant SNR gain on sub-carriers close to zero.
 R EFERENCES
   !
 

[1] K. K. Hung, P. K. Ko, C. Hu, and Y. C. Cheng, “A unified model for
  !
 
the flicker noise in metal-oxide-semiconductorfield-effect transistors,”
 Electron Devices, IEEE Transactions on, vol. 37, no. 3, pp. 654–665,
1990.


      
[2] E. Simoen and C. Claeys, “On the flicker noise in submicron silicon
  
!  MOSFETs,” Solid State Electronics, vol. 43, no. 5, pp. 865–882, 1999.
[3] O. Oliaei, “Noise analysis of correlated double sampling SC integrators
Fig. 9. Sub-band Wiener filtering of flicker noise, fc =2 MHz with a hold capacitor,” Circuits and Systems I: Fundamental Theory
and Applications, IEEE Transactions on, vol. 50, no. 9, pp. 1198–1202,
2003.
[4] C. Y. Chan, Y. S. Lin, Y.C. Huang, S. S. H. Hsu, and Y. Z. Juang, “Edge-
D. Flicker noise mitigation in OFDM systems extended Design for Improved Flicker Noise Characteristics in 0.13-um
RF NMOS,” Microwave Symposium, 2007. IEEE/MTT-S International,
Until now, we have evaluated prediction filtering of flicker pp. 441–444, 2007.
noise assuming that the transmitted signal, x(k) is Gaussian. In [5] E. A. M. Klumperink, S. L. J. Gierkink, A. P. van der Wel, and B. Nauta,
“Reducing MOSFET 1/f noise and power consumption by switched
this section, we evaluate the performance of prediction filtering biasing,” Solid-State Circuits, IEEE Journal of, vol. 35, no. 7, pp. 994–
(refer to Fig. 6) when the desired signal is an OFDM signal. 1001, 2000.
We consider the case when the transmitted signal x(k) is an [6] B. Razavi, “A study of phase noise in CMOS oscillators,” Solid-State
Circuits, IEEE Journal of, vol. 31, no. 3, pp. 331–343, 1996.
OFDM signal generated based on the IEEE 802.11a standard. [7] T. H. Lee, “The design of cmos radio-frequency integrated circuits, 2nd
Due to the nature of the of the flicker noise, we expect the edition,” Communications Engineer, vol. 2, no. 4, pp. 47–47, Aug.-Sept.
sub-carriers close to dc to be the worst affected. Figure 8 2004.
[8] M. Cao, Y. Zheng, and H. K. Garg, “A novel algorithm for dc
plot the SNR of the individual sub-carriers before and after offsets and flicker noise cancellation in direct conversion receivers,”
subband prediction filtering for fc = 2 MHz. The SNR of Communications Systems, 2004. ICCS 2004. The Ninth International
the input signal (measured over the entire band) is 0 dB. Conference on, pp. 441–445, 7-7 Sept. 2004.
[9] F. N. Hooge, “1/f noise sources,” Electron Devices, IEEE Transactions
We observe that there is substantial SNR gain on the lower on, vol. 41, no. 11, pp. 1926–1935, Nov 1994.
subcarriers. For instance, when fc =2 MHz, the SNR gain for [10] N. J. Kasdin, “Discrete simulation of colored noise and stochastic
1st subcarrier is about 4 dB. Note that the SNR gain decreases processes and 1/f power law noise generation,” Proceedings of the IEEE,
vol. 83, no. 5, pp. 802–827, May 1995.
for the subcarriers that the located away from dc , which is not [11] S. Haykin, Adaptive filter theory (3rd ed.), Prentice-Hall, Inc., Upper
surprising considering the fact that the effect of flicker noise Saddle River, NJ, USA, 1996.
reduces as the we move away from dc. [12] P. P. Vaidyanathan, “Multirate digital filters, filter banks, polyphase
networks, and applications: a tutorial,” Proceedings of the IEEE, vol.
E. Flicker noise mitigation through Wiener filtering 78, no. 1, pp. 56–93, 1990.
[13] P. Vaidyanathan, “Quadrature mirror filter banks, m-band extensions
In this section, we investigate the use of Wiener filtering in and perfect-reconstruction techniques,” ASSP Magazine, IEEE, vol. 4,
flicker noise removal. no. 3, pp. 4–20, Jul. 1987.

2596

View publication stats