News & Analysis

Compensation for Mixed-Signal Errors in 802.11a ZIF Receivers

Wolfgang Eberle, Boris Come, and Stephane Donnay; IMEC

10/31/2002 2:57 AM EST

Compensation for Mixed-Signal Errors in 802.11a ZIF Receivers
As IEEE 802.11a wireless LANs (WLANs) move from the R&D phase to shelves in consumer stores, cost pressures will continue to dramatically increase for designers. But, in some senses, designers are fighting an uphill battle on this front. 802.11a delivers a higher operating frequency, a complex orthogonal frequency-division multiplexing (OFDM) scheme, a 62-quadrature amplitude modulated (WAM) constellation, and a host of other complex RF requirements. Simply providing this functionality is difficult enough. Delivering it in a cost-competitive system is a daunting task, to say the least.

To overcome this price/performance challenge, designers can no longer rely on traditional RF techniques that revolve around enhancing superheterodyne architectures. On the contrary, designers need to implement zero-IF (ZIF) direct-conversion receiver architectures to even approach the price/performance requirements of today's 5 GHz WLAN designs.

The road to a ZIF transceiver, however, is paved with its own set of performance challenges. In particular, these designs are plagued by DC offset, mismatch, phase noise, and other challenges. Fortunately through mixed-signal compensation techniques, designers can account for these problems and deliver a more cost effective 802.11a radio to market.

Comparing Four Receiver Architectures
To illustrate the benefits that mixed-signal compensation in an 802.11a design, let's compare the performance of four receiver front-ends:

  • a superhet with two IF stages implemented as a discrete board design
  • the same superhet but with a system-in-a-package (SiP) RF front-end
    • li>ZIF with 5-GHz local oscillator (LO) for direct downconversion
  • a ZIF with a subharmonic mixer operating at 2.5 GHz for downconversion

Below, we'll use these radios to illustrate tradeoffs in automatic gain control (AGC) and DC offset compensation. Note that all four receivers were designed to meet the minimum signal-to-noise ratio (SNR) requirements for each modulation scheme at the minimum sensitivity level and cover a receive range from -85 to -30 dBm.

The Classical Superhet
The classical superheterodyne architecture is a trade-off between channel selection, sensitivity, and linearity ( Figure 1a). This results in multiple frequency translation steps for better separation of the tasks.


center>Figure 1a: Block diagram of a traditional super heterodyne receiver with two IF stages and a final digital upconversion.

Figure 1a shows a superhet with two IF stages followed by a digital downconversion using subsampling. Both the discrete and the SiP front-end contain a switchable RF variable gain amplifier (VGA ) [with a high and low gain setting] and an adjustable IF attenuator with 31 dB range in steps of 1 dB. Since there is no self-mixing, no DC path, and the downconversion to DC is performed digitally, this type of receiver does not suffer from DC offset saturation at the analog-to-digital converter (ADC). However, with respect to the large receiver dynamic range, one of the many cascaded blocks becomes easily saturated.

The ZIF Approach
ZIF receivers perform the downconversion with a single frequency translation.4 In these architectures, frequency translation, amplification, and filtering, are not separated very well (Figure 1b).


Figure 1b: Diagram of a ZIF 802.11a receiver.

In Figure 1b, the local oscillator (LO) appears at the same frequency as the received signal. Designing a mixer with a high LO-to-RF isolation is important since self-mixing results in a DC offset signal that is then largely amplified in the baseband section, hence saturating the ADC5. Thus, ZIF architectures must provide gain adjustment and DC offset correction capabilities to be successful in an 802.11a design.

Designers can reduce DC offset issues by developing a ZIF receiver with a subharmonic mixer that operates at 2.5 GHz. In this sub-harmonic mixing receiver design, the LO and RF signal do not appear at the same frequency.6 Thus, differential design techniques reduce the effect of static DC offsets in the baseband chain, leaving self-mixing-induced DC offset as the main source of problems.

Evaluating Compensation Techniques
Mixed-signal compensation techniques apply digital estimation (and compensation) to identify and reduce analog non-idealities. These compensation techniques can either relax front-end component specifications or increase system performance.

Compensation techniques can be employed to handle a host of analog imperfections. The most important include carrier frequency offsets (CFO), phase noise (CPN), large receive signal dynamic range, I/Q mismatches, and DC offsets.3,5,7

Conventional front-end design focuses on compensating analog problems directly in the analog domain. Increasing accuracy requirements and adaptivity for high performance, multi-band, and multi-mode receivers are in conflict with a self-controlled local analog solution.

On the other hand, a pure digital approach can, for example, not overcome the limited dynamic range in the front-end. A true mixed-signal solution benefits from the low signal processing cost in the digital domain and controls analog front-end elements only where absolutely unavoidable (Figure 2). Let's look at the control of two elements—AGC and DC offset compensation.


Figure 2: To perform AGC and DC offset compensation, designers must include a mixed-signal feedback circuit in their receiver designs.

AGC/DC offset estimation, compensation
Assuming that there are no analog RSSI signals, the estimation of signal strength and DC offsets is performed in the digital domain after A/D conversion. Both DC offset and too much signal gain result in saturation individually. However, when a weak signal requires high gain, this definitively leads to saturation when DC offset is present. At the beginning of the burst acquisition, both problems can be present at the same time. Hence, we need a joint AGC/DC offset estimation and compensation to separate the two effects and treat them appropriately.

For the entire receive dynamic range, this article explores the configuration space of the front-end based on an extended cascade analysis (Figure 3). The configuration space consists of all combinations of the gain and DC offset correction elements. This process identifies invalid receive regions due to saturation or insufficient SNR.


Figure 3: In a 5 GHz LO ZIF receiver with 15-dB LO-to-RF isolation, the useful range for the baseband VGA is very limited, since DC offset saturation occurs quickly.

The run-time AGC/DC offset implementation uses signal strength and offset estimates to address the configuration table shown in Figure 3 and retrieve the optimum configuration. However, saturation introduces a bias to the linear estimators, so we need a nonlinear approach.

The estimator overcomes saturation by classifying the receive signal into two saturation classes— NL-I for saturation due to DC offset and NL-II for saturation due to too high gain—and a linear case L. Classification is based on nonlinear operations such as threshold and sign comparisons. In cases NL-I and NL-II, nonlinear post-processing compensates for the biased estimate.

Estimation and compensation processes are applied sequentially in three steps compatible with the standardized preamble requirements, following a divide-and-conquer strategy.3 The first step resolves DC-offset induced saturation. The second handles gain-induced saturation, providing a non-saturated input to the third phase such that the linear estimator produces unbiased coefficients for the final gain and DC offset adjustment. Since the compensation only uses the optimum gain configurations, unpredictable cases do not exist (Figure 4). This ensures convergence to the optimum configuration in all cases.


Figure 4: In an 802.11a design, optimum gain for each RF input power is determined based on an extended cascaded analysis. This figure shows the results for high and low gain for a ZIF receiver with 24-dB LO-to-RF isolation.

Performance Results
Now let's compare the performance capabilities of the front-ends discussed above. During our analysis, we will derive the optimum gain configuration for each front-end and then compare the SNR performance of each front end. We'll also focus on the DC-offset problem in the ZIF architecture and look at the impact that the DC compensation technique discussed above will have on the ZIF radio.

Figure 5 shows a comparison of SNR performance for all four front ends at an optimal gain configuration.


Figure 5: Comparison of SNR performance for all four 802.11a receivers at optimal gain.

The comparison of SNR performance leads to three conclusions. First, the discrete digital-IF design should employ additional gain control to reduce the switching gap. Second, sub-harmonic mixing exhibits a performance gain for low receive signal levels, where any gain is especially valuable. Finally, improving the LO-to-RF isolation for the 5 GHz LO ZIF architecture leads to a significant performance gain.

Now that we've compared SNR, let's compare design cost. When considering design cost, we decided to analyze the front-ends with respect to maximum gain, gain range, and DC offset margin (Table 1).

Table 1: Comparison of Gain, Gain Range, and DC Offset Margin

LO-RF Isolation in dB Maximum available gain in dB Gain range in dB Margin between max. DC offset and ADC limit in dB
Discrete Superhet/Digital IF n/a 28 32 n/a
SiP Superhet/Digital IF n/a 21 31 n/a
Sub-Harmonic Mixing ZIF n/a 28 23 n/a
5 GHz LO ZIF 15
24		 37
37		 24
24		 2.1
11.1

As Table 1 shows, the superheterodyne designs require a large gain range in the last IF section. The ZIF architecture, on the other hand, requires a 9-dB higher maximum gain than the other architectures. Table 1 also shows that the sub-harmonic mixing ZIF front end features both moderate maximum gain and gain range requirements, and avoids DC offset problems, benefiting from the larger conversion gain and dynamic range of its mixer.

For the 5 GHz LO ZIF, the SNR depends nonlinearly on the LO-RF isolation of the mixer (Table 2). Increasing the isolation improves the SNR only in a moderate way. Still, applying a DC offset correction resulting in equivalent 24-dB isolation would result in 3.2-dB gain in SNR at the specified minimum sensitivity levels.

Table 2: Impact of LO-to-RF Isolation Increases

Offset Compensation
Figure 6 shows the additional baseband gain that can be applied to the 5 GHz LO ZIF receiver when applying DC offset correction. Except for the switching point between high and low RF gain, between 6 and 9 dB more gain can be applied without saturation.


Figure 6: A DC offset reduction equivalent to increasing the LO-to-RF isolation from 15 to 24 dB, allows significantly higher baseband gain.

Since all front-ends were initially designed to meet the minimum SNR requirements without DC offset compensation, we can also relax the front-end specifications and tolerate larger DC offsets, trading-off performance gain against reduced cost.

Final Thoughts
With cost pressures mounting, designers must fund ways to reduce costs in their front-end designs. The article above showed the performance differences of four receiver architectures for 5-GHz WLAN designs. Based on this analysis, it can be easily seen that a ZIF architecture sporting a sub-harmonic mixer provides the best solution when building 802.11a receiver front ends.

Author's Note: The authors would like to thank Hugo DeMan and Georges Gielen for their help with this article.

Editor's Note: This paper is based on a presentation made at the 2002 Communications Design Conference.

References

  1. IEEE 802.11a, High-speed Physical Layer in the 5 GHz Band.
  2. ETSI TS 101 475, BRAN: HiperLAN Type 2.
  3. W. Eberle, Jan Tubbax, et al., OFDM-WLAN Receiver Performance Improvement using Digital Compensation Techniques, IEEE RAWCON, Boston, August 2002.
  4. B. Razavi, Design Considerations for Direct-Conv. Receivers, Transactions on Circuits and Systems II, vol. 44, no. 6, pp. 428-435, June 1997.
  5. A. A. Abidi, Direct-conversion radio transceivers for digital communications, IEEE Journal of Solid-State Circuits, Vol. 30, No. 12, pp. 1399-1410, December 1995.
  6. M. Shimozawa, et al., A novel sub-harmonic pumping direct conversion receiver with high instantaneous dynamic range, in 1996 IEEE MTT-S Digest, pp. 819-822, June 1996.
  7. C. Muschallik, Influence of RF Oscillators on an OFDM Signal, Transactions on Consumer Electronics, vol. 41, no. 3, pp. 592-603, 1995.
  8. W. Eberle, et al., Automatic Gain Control for OFDM-based Wireless Burst Receivers", 5th OFDM Workshop, Hamburg, Germany, 2000.
  9. W. Eberle, et al., Flexible OFDM Transceiver for High-Speed WLAN, Proc. of Vehicular Technology Conference 99, vol. 5, pp. 2677-2681, September 1999.

About the Authors
Wolfgang Eberle is a senior researcher in the Mixed-Signal and RF Architecture Design Group (MiRA) in IMEC. He holds an MSEE degree from Saarland University, Germany, and is now finalizing his Ph.D. at KU Leuven, Belgium. Wolfgang can be reached at wolfgang.eberle@imec.be.

Boris Come leads the architecture design team in the MiRA group in IMEC. He holds an MSEE degree from ENSEEIHT, Toulouse, France. Boris can be reached at boris.come@imec.be.

Stephane Donnay is heading the research group on Mixed-Signal and RF Architecture Design in IMEC. He holds an MSEE and Ph.D. degree in Electrical Engineering from KU Leuven, Belgium. Stephane can be reached at stephane.donnay@imec.be.



Please sign in to post comment

Navigate to related information

EE Buzz DesignCon

Datasheets.com Parts Search

185 million searchable parts
(please enter a part number or hit search to begin)

Feedback Form