|
This article highlights some of the challenges faced in the design of integrated circuits for television receiver systems. These challenges are less common and can be more difficult to understand than those of some other communication receivers. They make television receivers (tuners & demodulators) among the toughest "radios" to build. In brief, the major design challenges are brought on by the wide TV input-signal bandwidth, the presence of potentially huge blocking signals, and the image and harmonic problems.
Traditionally, TV receivers have been designed using discrete transistors, accurately-tuned discrete inductors and capacitors, wide-band VCOs and high-voltage (15V and above) power supplies. These solutions don't translate very well into silicon.
Broadband Design
One major difficulty lies in the extremely wide television bandwidth, where input channels span the spectrum from nearly DC to 1GHz (actually about 48MHz to 864MHz). This poses some issues not normally dealt with in narrow-band RF design.
The first of these is the inability to use fixed narrowband (LC) circuits in the signal path " the RF path must be broadband, which implies using resistors, resulting in added noise. For example, whereas one might inductively degenerate a narrow-band LNA, a broadband LNA may require resistive degeneration, with an obvious penalty of added noise and a tougher third-order intercept point (IP3) versus noise figure (NF) tradeoff.
The second issue is the large number of distortion products that can fall onto the desired channel. Rather than a few adjacent channel power ratio (ACPR) and intermodulation (IM) distortion concerns, there are hundreds of distortion components that can fall onto a desired channel due to composite-triple-beat (CTB) and composite-second-order (CSO) distortion. The latter distortion effects may be relatively unfamiliar to some RF designers and are described below.
Common Distortion Terms
Let's start by looking at simple harmonic distortion, and pass a cosine of frequency ω and amplitude A through a weakly non-linear transfer function vo=a1vi+a2vi2+a3vi3. "Weakly" non-linear means that the coefficients a2 and a3 are close to zero, so the dominant effect of the equation is to create an output that is nearly a linear function of the input, with a gain of a1. But, since we're trying to build extremely linear circuits, even small non-linearities can have dramatic effects.
If vi=Acosωt, then (ignoring the power-of-2 term under the assumption of a well-balanced differential circuit):
Vo = a1Acosωt+a3(Acosωt)3= a1Acosωt+a3A3(cosωt)3= a1Acosωt+a3A3(¾cosωt+¼cos3ωt)
Compare the amplitude of the tone at 3ω (the third harmonic) with the amplitude of the tone at ω (the fundamental); this gives a measure called third-harmonic distortion (HD3), typically measured in dBc (-55 dBc HD3 means one has a third-harmonic that is 55 dB lower than the fundamental). In this case the result is: HD3≈¼a3A2/a1. That's approximate, since we assumed the size of the fundamental was just a1A and did not include the extra ¾a3A3 term that resulted from the distortion, which is all right since a3 is so small relative to a1.
One can perform the same calculations for second-harmonic distortion (keep the power-of-2 term in that case), and also for intermodulation distortion that results when two tones are sent through the non-linear transfer function. Figure 1 below highlights some of the terms and their sizes (again, the HD and IM values are shown as the delta between the fundamental level and the corresponding distortion-component level).
Figure 1
An interesting thing to note is what happens as the size of the input is reduced. Obviously, the fundamental at the output reduces one dB per dB reduction in the input. The terms due to the ()3 term in the non-linear function (i.e. the third harmonic and the IM3 terms) go down faster, however, at 3dB per dB, because of the A3 term next to the cos3ωt term in the equation above.
Since the fundamental drops at a rate of 1 dB per dB, and the "third" terms drop at 3dB per dB, the difference between them (i.e. the HD3 and IM3 measures) drop 2dB per dB reduction in input level. That's also clear since they have powers of 2 (for input amplitude) in their equations on the figure.
These are the most commonly considered distortion components, and one easily grasps two closely-spaced frequencies (say 820 and 830 MHz) causing "intermod" products at 810 and 840 MHz, or a single 270 MHz frequency causing a third-harmonic product at 810 MHz, i.e. a case in which a non-linearity would cause undesired inputs at 820, 830 and 270 MHz to all result in unwanted terms on a desired 810 MHz frequency of interest.
Lesser-Known Distortion Components
There is a second-order distortion phenomenon coined composite second-order (CSO) that causes distortion components at the sum and difference frequencies of two input tones. In narrow-band systems this is less of a concern, but in our wide relative-bandwidth system this can easily cause problems. For example, 300 and 510 MHz inputs to a non-linear system give CSO distortion tones at 210 and 810 MHz, both within our TV bandwidth (in addition to the tones created at 90 and 720MHz due to intermod). See Figure 2.
Figure 2
By working through calculations similar to those performed above for a single tone into a weak non-linearity, another distortion-related issue in wide-band systems emerges due to the combination of three input tones. Composite triple-beat, or CTB, combines sums and differences of three input-signal frequencies. One example is shown in the Figure 3. For three tones sized as were the tones in a two-tone intermod example, CTB gives distortion tones that are 6 dB higher than in the IM case.
For a wide-band system like television, the channel in the center of the band is particularly susceptible to CTB distortion products, there being 3N2/8 terms that can fall on the center channel of an N-channel system. The figure below shows tones at 350, 360 and 820MHz mixing to produce an unwanted term at 810MHz.
Figure 3
Thus, we've seen various ways in which unwanted blockers at 270, 300, 350, 360, 510, 820 and 830 MHz can all combine to distort a desired 810MHz band, and there are many other blockers in a wideband system that can do so as well.
LO-Harmonic Mixing Issues
Another impact that a broadband input has on a tuner design is on the choice of the first intermediate frequency (IF). Many integrated "radios" are best designed with a DC, or zero IF, which can greatly alleviate image-rejection requirements since the image is the same amplitude as the desired channel.
Alternately, one might choose a low-IF (several MHz, or perhaps a standard TV IF such as the European 36MHz) to allow inexpensive IF processing, though this places more strain on the front-end in terms of image rejection, since the image can now be a large nearby blocker. (Image-rejection issues are not unique to TV systems, so they won't be discussed further here.)
With either choice of IF above, the initial challenge is related to the harmonics of the square-wave local oscillator (LO) signal. These are silicon implementations, so we generate our LO output from a phase locked loop (PLL), and it is a square wave because to filter it into a sine wave across all necessary frequencies is not feasible. Since the LO is a square wave, it comprises a fundamental tone as well as decreasing-amplitude tones at the odd harmonic frequencies. These harmonics can mix higher-frequency channels to the desired IF.
|
