Estimation and Compensation o IQ-Imbalances in irect own Converters NRES PSCHT, THOMS BITZER and THOMS BOHN lcatel SEL G, Holderaeckerstrasse 35, 7499 Stuttgart GERMNY bstract: - In this paper, a new method or the estimation and the compensation o IQ-imbalances in direct down conversion receivers is presented. The considerations are based on a receiver structure that is developed or the simultaneous down conversion o up to our neighbouring carriers in UMTS base stations. The suppression o such a system must achieve 6dB at least. This reuirement is not ulilled by the analogue part and hence, an error estimation and compensation in the digital domain is necessary. In laboratory measurements using a W-CM signal, the suppression could be improved by 5.3 db to a resulting value o 75. db. Key-Words: - irect down conversion, suppression, IQ-imbalance. Introduction ue to high pressure towards cost reductions on the telecommunication market, a goal o the development is the integration o analogue parts, e.g. o the base station receiver, on SICs []. irect down converters or the single carrier reception in mobiles like [] are already existing. However or base stations, there is a demand to process up to our carriers simultaneously which reuires a higher bandwidth. ccording to the 3GPP speciications, e.g. a blocking intererer signal can occur in the receive band. Since the power level o this blocking signal can be much higher than that o the user signal, a high suppression must be ensured. Receiver Structure The receiver architecture comparable to [3]-[4] is shown in Fig.. It extracts the real and imaginary part rom a complex signal converting it directly down rom the RF domain to the base band. In the present case, the ield o application is the reception o a multi carrier W-CM signal in UMTS base stations. The antenna signal is iltered by a bandpass and ampliied by a low noise ampliier and a variable gain ampliier. The signal path is divided into the in-phase (I) and the uadrature (Q) paths, each containing a mixer, a lowpass and an ampliier stage. The LO ports o the mixers are driven by CW signals with a reuency eual to the centre reuency o the received multi carrier band. Thus, one o the mixing products occurs around the reuency while the other one occurs in the baseband. The LO input signals o the mixers must dier by a phase shit o 9. t the output o the mixers, the baseband signals are lowpass iltered and ampliied. terwards they are -converted. In the subseuent digital domain, the particular channels are separated and the error estimation and compensation is done. Fig. RF SIC 9 Q-Path I-Path LO Block diagram o a direct down conversion receiver. 3 Imbalance Problem Since in analogue IQ-demodulators, the I and Q paths cannot be built identically and the 9 -phase shiter is not ideal, gain and phase imbalances between the I and the Q signals occur. Thus the ollowing channel separation algorithm cannot separate the particular channels without mutual intererence caused by signals. The power level o these unwanted s depends on the occurring IQ-imbalances as described in [5]. n example in the case o the 3GPP blocking speciication is shown in Fig.. It represents the worst case o the problem in the considered base station receiver. In order to detect the channels correctly, the s and thereore the IQimbalances must not exceed certain limits. With the approach shown in Fig. 3, the dependency o the suppression on the gain and phase imbalance can be calculated.
RF domain IF domain Spectrum Blocking signal, level: -4 dbm LO - oset User signal, level: -5 dbm LO Freuency LO + oset Spectrum Image o the Blocking signal oset User signal, level: -5 dbm Freuency Fig. Illustration o the problem in the 3GPP blocking speciication. Fig. 3 pproach or the calculation o the suppression. The parameters k and k model the gain imbalances o the transmitter and receiver and the parameters p and p model the phase imbalances o the transmitter and the receiver respectively. The transer unction o the chain is. i i p k i p k cos sin (a) (b) i and are the inphase and uadrature input signals in the time domain and i and k are the inphase and uadrature output signals in the time domain. In the ollowing the actors / in (a) and (b) will be neglected or simpliication and only the demodulator or the receiver is considered. The suppression can be derived by eeding the demodulator with a CW signal. i cos t (a) sin t (b) With (a) and (b) the output signal can be calculated. i cos t (3a) k sin t p (3b) The coeicients or the wanted and the reuency can be calculated via ourier transorm o the signal in the time domain. The ourier coeicients are. d d k k p jk p cos sin p jk p cos sin (4a) (4b) d + is the ourier coeicient at the wanted signal and d - the ourier coeicient at the signal respectively.
With euation (4a) and euation (4b) the suppression can be derived. a Image d log aim d k cos log k age p k p k a Im age cos (5a) (5b) With euation (5b) the suppression can be calculated or given amplitude and phase imbalances o a demodulator. The result can be seen in Fig. 4. In order to obtain an suppression o 6 db or more which is reuired e.g. or the correct detection o the UMTS channels, the gain imbalance must not exceed. db and the phase imbalance must not exceed. degrees. 3 4 ntenna & uplex Filter c Receiver SIC I Q FPG Channel Separation (Channels & ) Channel Separation (Channels 3 & 4) Error Estimation & Compensation Error Estimation & Compensation Channel (I, Q) Channel (I, Q) Channel 3 (I3, Q3) Channel 4 (I4, Q4) Fig. 5 Block diagram o the complete receiver structure with channel separation and error correction.. Indirect Error Compensation The term "indirect error compensation" is used or a method to compensate the IQ demodulator errors by subtracting the error signal introduced by the IQ demodulator at the output o the digital down conversion (C). Because these errors are related to the insuicient rejection o the IQ demodulator they are correlated to the "reuency inverted" output o a down converted channel tuned to the channel. "Freuency inverted" means that the error signal within the wanted channel has an inverted rotation with respect to the channel signal. The transer unction o the indirect error compensation stage could be written as. 3 4 s out s * s in, wanted in, c (6a) i i a bi out in, wanted in, in, (6b) Fig. 4 Image suppression versus IQ-gain and phase imbalance. 4 Error Compensation and Estimation ue to the imbalance problem mentioned above, a method was developed or the estimation and correction o the signal errors ater the channel separation in the digital domain. The error estimation is done with a calculation o the crosscorrelation between two channels that are symmetric about C. Furthermore, the channel power o each channel is determined. From that, a complex correction actor is extracted that is used in the ollowing error compensation. The error compensation is carried out by weighted subtraction o the band rom the wanted band. Fig. 5 shows the block diagram o the resulting structure with channel separation as well as error estimation and compensation in the case o a our carrier receiver. ai b out in, wanted in, in, c a jb (6c) (6d) s out is the wanted complex output signal ater correction, s in,wanted is the wanted complex input signal beore correction and s * in, is the reuency inverted complex input signal. c is the complex scaling actor and i x is the real part and x the imaginary part o the signal. Fig. 6 shows a possible indirect compensator block diagram. For every wanted channel such a stage is necessary. The channel must also be available or the indirect error compensation. Fig. 7 shows a block diagram o a our channel multicarrier receiver with symmetrically spaced reuency channels. In this case the o one channel is the wanted signal o its counterpart, so that there is no need or implementing extra downconverters to obtain the signals needed or the compensators. Only two sets o scaling actors are needed or this case.
xy xxw c.5 xxi xxi or xxi xxw (9a) xy xxi c.5 xxw xxw or xxi xxw (9b) k.47 Im c p Re c (a) (b) Fig. 6 Block diagram o the indirect compensator stage. xy is the complex correlation coeicient between the wanted and the signal. xxw is the energy o the wanted signal seuence and xxi is the energy o the signal seuence respectively. i i and i are the inphase and uadrature components o the sample i and n is the number o samples taken or correlation. The second term in euation (9a,b) has been ound empirically. It has a signiicant contribution only i the power o the wanted and signal is about the same. Euation (a,b) is also ound empirically. The number o samples taken or the computations has been varied between one radio timeslot and / o a timeslot with no signiicant dierence. Fig. 7 receiver. Indirect compensation or a reuency symmetric our channel B. Error Estimation The error estimation unctionality is needed to estimate the unknown complex scaling actor. The error estimation is done by correlating the "reuency inverted" signal o the corresponding channel with the signal o the wanted channel ater the direct down conversion stages. The scaled complex correlation coeicient is used to compute the scaling actor used or indirect compensation. lso the amplitude and phase imbalance o the IQ demodulator could be calculated rom the scaled correlation coeicient. xy xxw n i, i j i ii, i i, i, n i i (7) i i,, (8a) wanted i wanted 5 Simulation Results In Fig. 8, a simulation result obtained with S rom gilent Technologies is shown. s an example, the suppression with and without error compensation is plotted versus the gain imbalance. The analogue part o the Receiver is included or the simulations. Compared to the suppression without compensation an theoretical improvement o up to 45 db can be achieved. s expected, the supppression is decreasing with increasing gain imbalance. For small imbalance values a saturation o the suppression occurs. The suppression is suicient or gain imbalances up to 4.5 db. xxi n i i i,, (8b) i Fig. 8 Simulated suppression with and without error compensation.
Fig. 9 shows the with and without compensator. Without compensator, the rejection is about 3 db which is the one o the uncompensated IQ demodulator. With compensator, the is totally masked by the noise loor. t least 3 db improvement is achieved in this case. Fig. 9 - -4-6 -8 - - -4-6 -8 - -3 - - 3 Main channel and, with and without compensator. 6 Measurement Results Measurements have been perormed to show the unctionality o the error compensation using a set-up shown in Fig.. Fig. PC Ring Buer I,Q digital FPG Channel Separation Evaluation o signal data Upper channel beore EEC Lower channel beore EEC Set-up used or laboratory measurements. Error Estimation& Compensation (EEC) Upper channel ater EEC Lower channel ater EEC The input o the FPG is connected to the digital data source which can be used or a CW as well as or a W- CM signal. The signal data in the upper and lower channel are read out rom the FPG beore and ater the error estimation and compensation part. Finally the data are post-processed in a PC in order to obtain the spectra o the particular channels. The irst measurement is perormed using a CW input signal with an amplitude imbalance o db that was.9 MHz above the receiver LO reuency o 95 MHz. In the error estimation algorithm, 89 data samples were averaged to calculate the cross-correlation and the channel power values. The spectra o the CW signal in the upper band and its unwanted in the lower band can be seen in Fig.. Spectrum / db - - -3-4 -5-6 -7-8 -9 signal at input and output beore compensation reuired limit ater compensation - -3 - - 3 Freuency / MHz Fig. Spectra o the CW signal in the upper band and its beore and ater compensation. The CW signal is passed through without a noticeable power level change. The is 4.8 db below the original CW signal beore the error compensation. ter the error compensation, the suppression has reached 7.3 db. That means that the error estimation and compensation algorithm has increased the suppression by 45.5 db. The second measurement was done with a W-CM input signal with a bandwith o 3.84 MHz and an amplitude imbalance o db. In the error estimation algorithm, again 89 data samples were averaged to calculate the crosscorrelation and the channel power values. The spectra o the W-CM signal in the upper band and its unwanted in the lower band can be seen in Fig.. Spectrum / db - - -3-4 -5-6 -7-8 -9 - signal at input and output beore compensation reuired limit ater compensation - -5 5 Freuency / MHz Fig. Spectra o the WCM-signal in the upper band and its beore and ater compensation.
The W-CM signal is also passed through without a signiicant power level change. The is 4.8 db below the original W-CM signal beore the error compensation. ter the error compensation, the suppression has reached 75. db. That means that the error estimation and compensation algorithm has increased the suppression by 5.3 db in this case. Fig. 3 shows a comparison between measurement and simulation o the suppression depending on the gain imbalance. gain a W-CM signal with 3.84 MHz bandwidth is used. In the error estimation algorithm, again 89 data samples were averaged to calculate the crosscorrelation and the channel power values. The simulation shows good agreement with the measured values even or large imbalances. 7 Conclusion This work presents a new method or the estimation and compensation o signal errors resulting rom imbalances between the I and Q paths in direct down converters. The unctionality is shown in S simulations as well as by a laboratory measurements. cknowledgement The author would like to thank Mr. Bergmann and Mr. Karthaus rom the tmel company who provided us with a receiver test SIC. Reerences [] U. Karthaus, F. Gruson, T. Bitzer, H. Vogelmann, G. Bergmann, N. lomari, K. Weese and. Pascht, Fully integrated, 4-channel, direct conversion SiGe receiver IC or UMTS base stations, IEEE European Microwave Conerence, Munich, Oct. 3. [].. bidi, irect-conversion Radio Transceivers or igital Communications, IEEE Journal o Solid-State Circuits, vol. 3, pp. 399-4, ec. 995. [3] T. Bitzer, U. Karthaus,. Pascht and K. Weese, Realisation o a SiGe-HBT irect own Conversion Receiver or UMTS Base Stations, IEEE EMO Symposium, Manchester, pp. 83-88, Nov.. [4] J. K. Cavers, daptive Compensation or Imbalance and Oset Losses in irect Conversion Transceivers, IEEE Transactions on Vehicular Technology, vol. 4, pp. 58-588, Nov. 993. [5] J. Jussila, J. Ryynänen, K. Kivekäs, L. Sumanen,. Pärssinen and K.. I. Halonen, -m 3.-dB NF irect Conversion Receiver or 3G WCM, IEEE Journal o Solid-State Circuits, vol. 36, pp. 5-9, ec.. Fig. 3 Comparison between measurement and simulation or the gain imbalance dependent suppression.