Design Article
The Need for Linear Circuits
Poojan Wagh
2/23/2009 2:41 PM EST
When I started working, it quickly became apparent that most of the problems with which I was dealing were not concerned with noise but with linearity. Everyone understands noise. However, if you want to differentiate yourself, you should understand linearity. This article explains why linearity is important and touches on how to improve linearity.
If you haven't already, you should at this point take a moment to review Christopher Bowick's detailed tutorial of RF design . He has laid a detailed technical foundation, and I don't think I can do better here.
Instead, I will focus on the trend of less filtering and more linearity in radio design, both as a differentiating factor for receivers and as a differentiating factor for analog/RF circuit designers.
On SNR
To be clear, noise and signal-to-noise ratio (SNR) are important. You should absolutely design your receivers to output a certain amount of SNR at some minimum input power (usually called the sensitivity level). However, most of these specifications are determined by the standard to which you are designing (GSM, GPS, WCDMA, 802.11) and really won't differentiate your product.
I am not saying that SNR is optional; I am saying that it is mandatory, and therefore won't cause you or the products you design to be uniquely valued.
Interference
It turns out that there's another problem that receivers need to deal with: interference. By interference, I mean any waveform that appears at the antenna other than the signal intended to be received. Typically, this waveform is a signal transmitted by a third party bound for another receiver, but received by your antenna.
Engineers tend to distinguish between noise and interference, noise being random and interference having some predictable structure. However, the impact on the receiver is the same: both degrade bit error rate. We generally model the interference as a pure sinusoid riding on top of our signal. It is typical practice for RF designers to model signals as either a single tone (constant amplitude) or as two tones (variable amplitude) to capture linearity effects. Most RF designers don't bother to model the actual binary modulation. Typically, a system designer takes the statistics of the binary modulation and translates them into a single or two-tone specification. We can classify the modulation by where in the frequency-domain it exists. The plots below show different cases of interference, as one would see on a spectrum analyzer:
The cases when the interference is much larger than the desired signal (as shown in the illustrations) usually limit performance.
In the commercial case, direct in-band interference is unusual. The FCC licenses bands of spectra to ensure that direct interference does not occur. In unlicensed bands (802.11 for example), the system designers usually do a good job of ensuring that a reasonable amount of in-band interference can be tolerated while maintaining low bit error rate.
In the defense space, all bets are off, but most government agencies will agree on some reasonable specification that they expect. Of course, if you can beat their specification, you will be rewarded.
Often, the case where interference is in an adjacent band is a bigger issue than the direct interference case. The RF circuits themselves display a nonlinearity (usually modeled as a polynomial) that can cause this out-of-band interference to appear in-band. This translation can occur with both third order (IM3/IP3) and second order (IM2/IP2) nonlinearities.
Finally, harmonic interference can be a problem when the receive mixer translates the harmonics of the LO in addition to the fundamental (desired) component. This is very prevalent with switch-mode mixers. In addition, it can occur if the LO has harmonic content.
The figure below shows several cases of out-of-band signals mixing in-band:
The solution for several decades has been to place filters at strategic places in the receiver. The following diagram illustrates these strategic points and the qualitative effects of filtering. The green spectral lines show the hypothetical case where the filter ahead of the circuit is omitted. The red spectral lines show the case where the filter is included. Note, however, that the distortion is not to scale. If I reduce interference by 1 dB, my IM3 goes down by 3 dB. However, to illustrate the case, I show IM3 only going down proportionally to the interference.
The idea is that the circuits induce nonlinearity. As a result, one should place filters ahead of these circuits to reduce the unwanted interference. Of course, filtering isn't an absolute solution to the problem. No filter has infinite selectivity—the ratio (expressed in dB) of gain in the pass-band to gain in the stop-band. Filters either decrease signal power in the pass-band (thus increasing noise figure) or they pass some portion of the unwanted signals in the stop-band. These deficiencies have been around for years, and have generally been accepted as status quo.
Comments
frankmlinar
2/25/2009 7:18 AM EST
Interesting article. Unfortuantely, it is too one-dimensional. As a cosite engineer, I am all too aware of the various interference mechanisms and potential solutions. To focus exclusively on the cost/linearity tradeoff ignores other issues such as size and power consumption. In my opinion, the optimization space should include as a minimum size, weight, power, performance, and cost. The weightings of these factors will change depending on the application. Spectrum planning and software can be additional factors.
Sign in to Reply
W8GT
3/4/2009 10:23 AM EST
A very lucid presentation of the problem. Without improved linearity in the receiver, RF and IF filters are here to stay. Simply increasing the current in the amplifiers is not practical. We need sophisticated linearizing techniques. RF power amplifier designers have gone a long way down this road. Now it is time for receiver designers to do the same.
ADI, LTC are you listening?
Sign in to Reply
Derek Wood
3/6/2009 4:44 AM EST
Linearity does not stand alone when it comes to blocking performance. It is only one of the defining parameters. Others are
1. VCO phase noise
2. Architecture
3. Environment
4. Modulation
5. Protocol
(non definitive list)
In the most recent product I've been working on we have a receiver and transmitter that can both work on only 3mA at 1V. You don't get linearity without supply current. i.e. if an LNA is fed with a -10dBm input signal but only has 0.3mW power budget you WILL limit with gain above 5dB.
The solution? get a SAW filter on the front end, this does not need power but cuts down the likely interferers by 40dB. LC filters just won't do it here.
VCO: If you have poor phase noise you will suffer reciprocal mixing, if you want to reduce phase noise you will have to pour more current into the VCO, all other things being equal. A frac-N PLL may help by reducing the close in phase noise but beyond +/-1MHz you're not going to benefit, indeed frac-Ns are sometimes used to allow for a poorer VCO (low power budget) thus making the reciprocal mixing worse.
Architecture? try jamming a direct conversion receiver then comparing it against a superhet based design. The direct conversion receiver will fail very very easily, the superhet will carry on and on unless you are lucky and hit it's reception channel.
Environment: If you know where or how jammers are going to hit you you can take advantage, i.e. put a notch in the middle of the mobile phone band, are your jammers very short lived (i.e. a swept radar pulse) in which case spread your channel over time and include redundancy.. which brings us to:
Modulation: Can you afford redundancy, most systems include a degree of redundancy, the more you can build in the more interference resistant you will get, do you need complex or simple modulation schemes, the more complex the scheme the more demand is placed on the receiver demod, which means greater silicon area (cost) or more current (less portable battery)
Protocol: can you hop round the currently defective channel?
Linearity, yes you want as much of this as you can get, which is why when you construct a system design you don't only consider the noise along the receive path but also the compression point, but it is only one part of a much more complex picture.
Oh and finally noise figure, but that's a 'given' in the design of all receivers.
Sign in to Reply