DC offset cancellation in a zero if receiver

[0001] The invention relates to a method and apparatus for correcting DC offset problems found in high-speed transceiver systems.

[0002] Conventional transceiver systems commonly must switch back and forth between transmit and receive modes. As the speed of data transmission increases, the time allowed for switching between the Transmit (Tx) and Receive (Rx) modes becomes smaller. One conventional type of receiver is known as a Direct Conversion zero intermediate frequency (IF) receiver. In this type of receiver a local oscillator is tuned to the carrier frequency of the incoming signal. These types of receivers commonly have multiple stages in which the incoming signal is down converted and processed using a local oscillator (LO) circuit. These IF type receivers create in phase (I) and 90 degrees out of phase quadrature (Q) signals from the received signal. Large DC offsets are produced in the down converter outputs of Zero IF receivers due to LO leakage at the RF ports of the down converters. Additional DC offsets exist along the I and Q paths that include low-pass channel filters and automatic gain control (AGC) circuits with large gains. Therefore each of these DC coupled stages introduces a DC offset error into the signal. In an orthogonal frequency division multiplexing (OFDM) system, a difference between the local oscillator (LO) frequency and the incoming signal frequency causes DC offset errors within the system to profoundly degrade the SNR after demodulation. These unwanted DC errors may also cause the amplifiers used in the I and Q branches to saturate. Once the amplifier is in a saturated state, the received data signal cannot be processed and amplified correctly so the received data signal is lost.

[0003] Prior art attempts to deal with the above problems have been only semi-successful. Stroet et al.'s article entitled “A Zero-IF Single Chip Transceiver for up to 22 Mb/s QPSK 802.11b Wireless LAN” shows that these offset errors may be reduced or settled in 25 microseconds. The reduced DC offset is still too high for OFDM systems and takes too long to settle. As mentioned above, with an increase in data speeds, this prior art system is not useable in today's transceiver environments.

[0004] All known prior art techniques used to reduce these DC offset errors have drawbacks in one form or another. For example, AC coupling signals with high frequency cut off values may reduce response time, but has an unacceptable signal to noise ratio or an unacceptable effect on the signal itself Further, automatic gain control is also only useable when the DC offset level is very small. The end result is that these DC offset values can not be reduced in an acceptable amount of time.

[0005] The instant invention uses an automatic gain controller AGC and a digital signal processor along with zero IF transceiver circuitry to create I and Q signals to process and remove DC offset signals in acceptable time periods. Using a combination of techniques such as AC and DC coupling, automatic gain control, and digital signal processing, the DC offsets are removed to insignificant levels. The main features or steps of the invention are: 1) dynamically changing the cut off frequency of an AC coupling stage; 2) computing a DC signal error over a time period in which the I and Q signals complete a single or multiple cycles, 3) subtracting the estimated DC errors from the I and Q signals, and 4) high-pass filtering the resultant signal so that residual DC errors are removed. The implementation and timing of these steps allows for DC offset control that is far superior to prior art systems.

[0006] In a second embodiment a D/A converter is used in each of the I and Q branches to significantly reduce the DC offset, and then steps 2–4 as mentioned above are followed. The DSP does a coarse DC correction with the D/A converter, and maintains a list of correction values for all combinations of antenna diversity, LNA amplifier gains and AGC gains. This ensures that DC offset correction values are maintained for all combinations of antenna diversity and low noise amplifier gains. In the second embodiment, the DSP controls the system so that the DC offset error is reduced when AGC gain is reduced.

[0007] In a third embodiment, the transceiver frequency error is further removed in the receiver by changing the voltage controlled oscillator crystal (VCXO) frequency that is used for the local oscillator (LO) frequency synthesizer. This also ensures that the effect of the DC offset error is minimized after demodulating an OFDM signal.

[0012] Using a combination of techniques such as AC and DC coupling, automatic gain control, and digital signal processing, the DC offsets are removed to insignificant levels. The main features or steps of the invention are dynamically changing the cut off frequency of an AC coupling stage; computing a DC signal error over a time period in which the I and Q signals complete a single or multiple cycles, subtracting the estimated DC errors from the I and Q signals, and high-pass filtering the resultant signal so that residual DC errors are removed. The implementation and timing of these steps allows for DC offset control that is far superior to prior art systems.

[0013] In a second embodiment a D/A converter is used in each of the I and Q branches to significantly reduce the DC offset, and the steps as mentioned above are followed. The DSP does a course DC correction with the D/A converter, and maintains a list of correction values for all combinations of antenna diversity and LNA amplifier gains. This ensures that DC offset correction values are maintained for all combinations of antenna diversity and low noise amplifier gains. In the second embodiment, the DSP controls the system so that the DC offset error is reduced when AGC gain is reduced.

[0014] In a third embodiment, the transceiver frequency error is further removed in the receiver by changing the voltage controlled oscillator crystal (VCXO) frequency that is used for the local oscillator (LO) frequency synthesizer. This also ensures that the effect of the DC offset error is minimized after demodulating an OFDM signal.

[0015] FIG. 1 shows a diagram of the transceiver 10 as a zero intermediate frequency radio device according to the present invention. The transceiver 10 comprises a receive branch Rx and a transmit branch Tx. A transmit power amplifier 14 is coupled to a Tx/Rx switch 13. The Tx/Rx switch 13 is coupled to an antenna 11. The transmitting branch is well known in the art and is not shown in detail here. The receiver branch further includes a variable gain low noise radio frequency amplifier (LNA) 15 that is coupled to the Tx/Rx switch 13. The LNA 15 amplifies an input signal that corresponds to an incoming radio frequency signal that is received by the antenna 11.

[0016] The output of the LNA 15 is coupled to a frequency down converter 34 for down converting the radio frequency signal to a zero intermediate frequency (IF) signal. The present invention employs a quadrature frequency down converter. The frequency down converter 34 which contains mixers 16 and 17 in respective quadrature and in-phase mixer paths that provide filtered and amplified quadrature signals Q and I. The frequency down converter 34 further includes controllable AC couplers 22 and 23, and channel filters 24 and 25. The AC couplers 22 and 23 are coupled between the mixers 16 and 17 and the zero IF amplifiers filters 24 and 25.

[0017] Control signals for the automatic gain control AGC 31 are provided by a DSP baseband processor 30. The baseband processor 30 contains processing means for providing cut-off frequency control signals and signals to the AGC unit 31. Signal line 35 controls the automatic gain controlling data, while signal line 36 contains frequency cutoff information. Also shown is a State Machine 9. The State Machine 9 is connected to capacitors 22–23 and 32–33. The State Machine is used to change the AC coupling frequency from 10 MHz to 500 KHz as will be explained in greater detail below. The operation and functions of the DSP 30 will also be described in detail with reference to FIG. 2 below.

[0018] In another embodiment, the baseband circuit 30 further employs analog to digital converters 20 and 21 for canceling the DC offset in the quadrature signals I and Q. The sampled I and Q DC correction signals are supplied by the digital signal processor (DSP) 30.

[0019] The transceiver 10 further comprises a PLL 19 for generating local oscillator signals for the receive branch Rx and for the transmit branch Tx. As is well known in the art, the PLL comprises a voltage controlled oscillator (VCO), a loop filter, and an integrator. A reference oscillator signal, as shown in more detail in FIG. 3, is supplied to the PLL. In order to generate the I and Q signals, a ninety degrees phase shifter is used in conjunction with the LO signals that are fed to the mixers 16 and 17. The transceiver 10 does not use the D/A converters in the first embodiment but does employ them in a second embodiment.

[0020] As described above, the problems with prior art transceiver systems are the unwanted DC offset values that are produced by the system components.

[0021] FIG. 2 shows a timing diagram according to one preferred embodiment of the invention. This timing diagram shows the durations of the AC and DC coupling stages necessary to reduce the DC offset errors. Also shown in FIG. 2 are the lower cut off frequencies of the filters during these AC coupling stages.

[0022] The invention implements both AC and DC coupling in the receiver I and Q base band paths. Temporary AC coupling is used to remove DC offsets that could otherwise saturate the receiver outputs due to the large gain in the base band paths from the down converters outputs to the I and Q outputs. The AC coupling is implemented as a cascade of one or more first order high pass filters 24 and 25, with a particular lower 3 dB cut off frequency (f lower).

[0023] With reference to FIG. 2, upon entering the receive (RX) mode at time zero, (step 1) the AC coupling cutoff frequency (f lower) is momentarily kept at 10 Mz for 0.15 usec. This is done automatically by a state machine 9 in the receiver. This quickly removes all the DC offset in the receiver I and Q base band paths. After 0.15 microseconds (f lower) is automatically reduced to 500 kHz, and remains at this cut off value until the DSP removes the AC coupling and introduces DC coupling. With the 500 kHz AC coupling, DC offset changes (with AGC gain changes) are quickly removed before the signal is sampled or the next AGC iteration. It is during this stage of step 1 that automatic gain control is performed on the signal. The gains of amplifiers 26,27,28 and 29 are changed by the DSP to adjust the IQ signals to a desired level at the AID input. FIG. 2 shows 3 distinct time periods of adjustment, however more periods could be used as necessary. In the example shown in FIG. 2, the DC coupling (step 2) is switched on at 1.05 microseconds.

[0024] The DC coupling is actually an AC coupling with a very low value of (f lower) which is less than 100 Hz. With such a low cut off frequency, it may be considered DC coupling even for long IEEE802.11a data packets that may be up to 5–6 milliseconds in duration.

[0025] It should be noted that whenever the LNA 15 gain is changed, there is a change in logic level at the antenna select input, and the state machine 9 changes (f lower) to 10 MHz for 0.15 microseconds and then returns (f lower) to 500 kHz.

[0026] It should also be noted that when the state machine 9 changes (f lower) from 10 MHz to 500 kHz, there is a step in the I and Q DC levels that may be as large as the peak signal level. This step quickly decays away to very low levels within about 0.8 microseconds. Simulations show that with 500 kHz AC coupling for IEEE802.11a, if a moving average signal power estimate is computed by the DSP 30 (averaged over 0.8 microseconds), then the error is within 2 dB after the first 0.8 microseconds of the RF burst. For a coarse signal level estimate, a 150 nanosecond averaging window (samples at 40 or 80 MHz AID) is sufficient.

[0027] When the DSP 30 changes the receiver base band AGC gain in step 1, it must wait for up to 300 nanoseconds before sampling the I and Q signals for computing the signal power. This is because of the transient settling of the receiver AGC DC levels that takes less than 300 nanoseconds to settle for a 500 kHz AC coupling. For small changes in AGC gain settings, the transient settling time is less than 300 nanoseconds. All changes in LNA and AGC gain settings should be done in order to get the proper signal level and allow the DC errors to be removed before making the decision for the next gain setting.

[0028] After finally adjusting the receiver AGC gain, the DSP 30 changes the receiver paths to DC coupling (step 2). When this is done, there is a small change in the I and Q DC levels, and it can not be avoided. It is less than about −5 dB relative to the peak signal level and is due to the AC coupling acting on the signal itself (not related to the actual DC error in the circuit). This DC error remains nearly constant during the rest of the receive burst that may be up to 6 ms long. In this preferred embodiment, this static DC error should be removed digitally by the DSP 30, only after the I and Q A/D conversion takes place. This will ensure not to degrade the signal to noise ratio (SNR) after the fast Fourier transform (FFT) in the receiver, especially when there are large relative frequency offsets between the Transmitting and Receiving modes in the transceivers.

[0029] In step 3 of the preferred embodiment the DSP 30 computes the DC offsets in the I and Q remaining parts of the signal. The average values of the I and Q signals are calculated by the DSP 30. The computed average DC offsets should then be subtracted from their respective signals for the rest of the packet. It is important to calculate the DC offset error at this point while the signal is DC coupled, so as to get an accurate indication of the offset error. After this subtraction, a first order high pass filtering should be done on the following I and Q signals digitally by the DSP 30, which represents step 4 of the preferred embodiment. This process significantly improves the SNR of the signal. This is because the estimated DC offset is not the true DC offset when the transmitter-receiver frequency offset is present. A residual DC error remains and it must be removed. Simple high pass digital filtering in the DSP 30 is sufficient. A lower cutoff frequency of 1 kHz is optimum for this digital filtering.

[0030] This process as shown in FIG. 2 allows the DC offset error to be reduced to acceptable levels within 8 microseconds. In this preferred embodiment it is assumed that the DSP 30 takes 150 nsec to compute the signal power and program the AGC. Actual time for worst case AGC setting will depend on the exact processing delay of the DSP, and the total number of AGC set iterations, and whether or not antenna diversity is used.

[0031] In another preferred embodiment, instead of computing the DC error in the DSP and subtracting it, the DSP can instead ramp down the value of (f lower) the AC coupling cut off frequency, from 500 kHz to less than 100 Hz over about 4 microseconds. This avoids the sudden step in DC offset that is associated with abruptly changing the cut off frequency.

[0032] FIG. 3 shows a third embodiment of the present invention, wherein a frequency adjust signal is applied to the voltage controlled crystal oscillator 50 to change the frequency of the LO. A signal (F adjust) is applied to the voltage controlled crystal 50 from the DSP controller. The phase locked loop contains conventional components such as a charge pump multiplier 51, a low pass filter 52, an integrator 53 and voltage controlled oscillator 54. The function of this circuit shown in FIG. 3 is to ensure that the frequencies of the local and received carriers are the same. When the frequency error is significantly reduced, even a large DC offset does not degrade the SNR after demodulation of the OFDM signal. Therefore the DC estimation and subtraction, and removal of residual DC error, is not required.

[0033] FIG. 4 shows one embodiment of how the automatic gain control circuit (AGC) and AC coupling may be implemented. This feedback circuit contains two amplifiers 60 and 61 with respective gains of G and A respectively. A low pass filter (integrator) 62 is also added after the feedback amplifier 61. This type of connection allows the transfer function from input to output to be frequency dependent. By changing the gains of the amplifiers 60 and 61, the cutoff frequency of this circuit may be varied. The −3 db lower cutoff frequency of the AC coupling is 2piAG. The product of AG must be maintained constant when changing the signal path gain G, in order to keep a constant cutoff frequency. In this manner automatic gain control may be implemented while keeping a constant lower cut-off frequency. As mentioned above the AC coupling provided by this circuit may be effectively changed to DC coupling by making the value of A very small, so that 2piAG is less than 100 HZ.

[0034] In view of the foregoing, it will be evident to a person skilled in the art, what various modifications may be made in the embodiments given, such as digital signal processing, gain control, channel filtering, and reduction of cut off frequencies. Further the invention is thus not limited to the examples provided.