SI
SI
discoversearch

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor.  We ask that you disable ad blocking while on Silicon Investor in the best interests of our community.  If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Technology Stocks : C-Cube -- Ignore unavailable to you. Want to Upgrade?


To: DiViT who wrote (35689)9/4/1998 2:09:00 PM
From: BillyG  Respond to of 50808
 
Where did ATI get its MPEG-2 expertise?
msdmag.com (click on the August Issue and then on the article by Kovacevic)

Scroll to the bottom of this article.

Mastering MPEG-2 Clock Recovery: Part II

by Branko Kovacevic

Last month (August Multimedia Systems Design) we discussed the role of
voltage-controlled crystal oscillators (VCXOs) in the recovery of the system clock in an
MPEG-2 receiver, and we examined a discrete model of a VCXO. This month, we'll cover
some main characteristics of VCXOs, and look at a unique way to design a VCXO with
discrete components.

Main characteristics of VCXOs

A voltage-controlled oscillator is built around a quartz crystal whose oscillation frequency
can be varied by applying a voltage control signal. In many MPEG-2 receivers, this signal is
usually generated as a pulse width modulated (PWM) or pulse density modulated (PDM)
signal with 12-bit resolution and a programmable frequency of the PWM carrier from
200kHz to 2MHz. With a passive or active RC low pass filter with a bandwidth from
20kHz to 250kHz (depending on the carrier frequency), this signal is transformed to the
control voltage Vc. A higher carrier frequency of a PWM signal is desirable due to lower
values of the capacitor used in the RC filter (in the nF range). Figure 1 shows a typical
structure and f = g(Vc) curve, while Figure 2 shows possible ranges of tolerances (ñ50,
ñ100, or ñ150ppm). Basic characteristics of a VCXO are:

 Central frequency f0 at Vc = Vc_max / 2 = 2.5V is 27.000000MHz

 Range of control voltage Vc is usually 0V-5V or 0.5V-4.5V to avoid varactor diode
nonlinearity

 Minimum frequency deviation fmin = -50ppm of f0 = -1,350Hz

 Maximum frequency deviation fmax = +50ppm of f0 = +1,350Hz

 Operating temperature range 0§C-70§C

 Passband of the control signal Bc - 3dB = 15kHz

 Tolerances of the frequency pull range are typically 3:1, 2:1 for selected and more
expensive VCXOs. For a 50ppm VCXO, a possible frequency deviation could be 3 x f or
4,050Hz

 Aging effect influences f = g(Vc) characteristic at five years, minimum

 Minimum deviation of the central frequency, f0, is defined as the difference between the
ideal (27.000000MHz) and real central frequency at Vc = Vc_max = 2.5V, and is usually
ñ25ppm (could be a negative value). A lower value (e.g., ñ20ppm) increases the price of
the VCXO

 Linearity of the frequency-voltage characteristic is usually 10%. A more linear VCXO
(5%) is usually more expensive

 Incremental linearity is defined as fmax / fmin = 3:1. VCXOs with better incremental
linearity (2:1) are more expensive

 Maximum phase jitter is 100psp-p

 Control voltage range for a 5V rail-to-rail VCXO is 0.1V-4.9V

 Control voltage range for a 3.3V rail-to-rail VCXO is 0.1V-3.2V.

Increasing the working temperature of the VCXO causes a shift of the characteristic f =
g(Vc) toward negative frequency deviations (see Figure 3), and in effect reduces the range
of frequency change to 20ppm, maximum. For this reason, a ñ50ppm range is not usually
used in MPEG-2 decoders; i.e., a ñ100ppm range is more common.

The frequency-voltage sensitivity variation over control voltage range shown in Figure 4
determines a maximum gain Kr for a stable PLL. It is defined as a gain where S is maximum.
Sometimes, a behavior of the VCXO is determined by the best straight-line approximation
of its frequency-voltage curve and is used in the model.

The pull range (maximum deviation of the frequency) has to be wide enough to compensate
for all sources of frequency errors such as tolerances, stability of the quartz crystal,
temperature, and aging. Increasing the pull range can be achieved with varactor diodes with
a greater range of capacity change, which also increases the price of the VCXO. To keep
good stability, low TTC, and low aging, a crystal with AT cut with fundamental frequency is
required. AT cut limits frequency deviation to ñ100ppm, maximum. For larger ranges, a
VCXO for first or third overtone with a frequency multiplier is required, which again
increases the price of VCXO. Usually, a choice of VCXO is a price/performance
compromise. In MPEG-2 receivers, the best choice is a VCXO for a fundamental
frequency with a ñ100ppm pull range.

Discrete implementation of the VCXO requires soldering of the crystal oscillator, varactor
diodes, and control electronics to the PCB, and exposition of all components to
high-temperature stress. A trimming capacitor is very often required, which is impractical in
mass production. On the other hand, an integrated VCXO is encapsulated in a miniature
metal case with the laser trimming of the load capacitance and has pretty much guaranteed
performance specs. Instead of a chip with $0.1 worth of crystal, a selected crystal can be
used, where the pull range is measured in only three points: at minimum, one-half, and
maximum control voltage, so that the total pull range is ñ50ppm, minimum.

Figure 5 shows the influence of the drive level to the relative change of the resonant
frequency of the crystal oscillator. The amplitude of the mechanical vibrations of the crystal
resonator increases with an increase in the excitation current. The power of excitation is
represented as Pc = R1 x Ic2, where R1 is the dynamic resistance of the crystal. An
increased excitation current leads to mechanical distortions of the resonator and vaporizing
of disc electrodes. The maximum limit of the excitation power is 10mW. The relative power
of the oscillation between the L1 and R1 elements of the crystal unit depends on the Q factor
as Prel = Q x Pc. With a drive power of 1mW and a Q of 100,000, it is 100W. Another
important influence of the increased drive level is the increase in the resonant frequency. The
recommended drive level is 0.1mW, and the minimum drive level is 1nW. Variation of the
resonant frequency with time has two components: short-term stability up to 10s, which
depends on the design and mechanical properties of the crystal resonator; and long-term
stability of a few days, months, or years, known as aging. The purity of the crystal structure,
stability of the inert gas (e.g., neon), and stability of the metal case against gas leakage affect
this property. Aging is a dominant effect in the first 1% of the useful life of the crystal unit
(see Figure 6).

Aging can be passive due to storage before usage of the crystal unit, and active as a
frequency shift under normal working conditions (i.e., during permanent oscillation). A
typical rate of aging is 0.5ppm per year. For low aging and good long-term stability, a low
drive level is desirable (10æW is usual). On the other hand, for good short-term stability and
low phase noise, a higher drive level is required, around 500mW. Artificial or accelerated
aging is usually applied to eliminate long-term effects and achieve good short-term stability at
drive levels less than 10mW.

A VCXO design with discrete components

Figure 7 shows a model of the crystal oscillator and the effect of a serial or parallel load on a
resonant frequency. The equivalent circuit consists of a dynamic inductance L1, which
models the mass of the oscillating disc whose elastic properties are modeled with the
dynamic capacitance C1. Dynamic resistance R1 models the finite amplitude at the
resonance or a finite Q value. Multiharmonics oscillations (overtones) and spurious
resonance can be modeled with additional parallel branches with RLC elements. At
frequencies higher than 100MHz, parallel conductance G0 models aging effect, erosion of
electrodes, and gas leakage through contacts. Parallel capacitance C0 models the small
capacitance of contacts. CL models the capacitance of a loading varactor diode with DC
reverse polarization voltage, which is a control voltage.

Figure 7 shows that serial load affects only the serial resonant frequency where Im[Z] = 0,
used in the actual design. Parallel load affects only the parallel resonant frequency (Im[z] ’
). Figure 8 shows an implementation of the VCXO with discrete components only. The
oscillator port in the Bt9104/Bt867 video encoder is used to place a crystal unit.
Option-ally, the oscillator port at the MPEG-2 transport demultiplexer can be used. The
video encoder operates in the master mode, generating horizontal and vertical sync pulses in
the system. The MPEG-2 video decoder is a slave. A control PWM signal, originating from
the MPEG-2 transport demultiplexer, is applied at the input of the RC passive low pass
filter. At the output of this filter, a control voltage is generated as a reverse polarization of the
varactor diodes. Toshiba's 1SV217 diode is used in this implementation.

Figure 9 shows a capacitance change of varactor diodes with a change of reverse voltage.
In the 0V-5V range of control voltage, there is a 52-20pF change of varactor diode
capacitance. With a serial load of 56pF, the total change of load capacitance that the crystal
unit "sees" is 15-27pF. Figure 10 shows a temperature effect at various voltages of the
reverse polarization. The relative capacitance change ratio has a minimum at room
temperature and is higher if the control voltage is smaller. A parallel resistor of 2.2M helps
start oscillation at power up. Crystal units from FOX, Bomar, and Pletronics were used with
stability of ñ50ppm, frequency tolerance of ñ30ppm, drive level of 1mW, aging of ñ5ppm
for one year, and dynamic resistance R1 of 20. The nominal declared capacitive load is
20pF. At the SMD crystal case, a maximum frequency deviation is 3-6ppm/pF, yielding a
total frequency pull range from ñ21ppm to ñ42ppm. For a larger pull range, a larger load
capacitor of 220pF instead of 56pF has to be used. This, however, increases phase noise
and decreases the spectral purity of the recovered clock. A better way to achieve a
ñ100ppm pull range is to do a custom design of the quartz crystal unit with adequate
dynamic capacitance that allows a frequency deviation of 15ppm/pF for a total pull range of
210ppm or ñ105ppm.

Figure 11 shows the frequency-voltage curve of the implemented circuit. At a PWM value
of -2,048, a zero deviation is achieved, and at 3,687 increments the frequency deviation is
200Hz. The selectivity or slope of the curve is Spwm = 190Hz / 5,735 increments, yielding
Kvcxo = 0.03313Hz/ increment, which was used in the discrete model of the VCXO.
Voltage sensitivity is 397Hz/V. From Figure 11 it is clear that the curve is asymmetric
because:

 At Vc = 2.5V, f = 180Hz, not zero

 At Vc x 1V the curve is changing (reducing) slope due to much smaller change of the
varactor diode capacitance

 Total frequency deviation is 1,000Hz + 750Hz = 1,750Hz or 64.8ppm and is asymmetric
as +37.1ppm | -27.7ppm.

Figure 12 shows the measured jitter of the NTSC color subcarrier of encoded analog video
at the output of the Bt9104 video encoder with a color burst frequency error of 20.7Hz.
Measurements were made with a Tektronics VM700 analog video analyzer. Most of the
color burst jitter is located in the range of ñ12ns. Color burst accuracy must be better (less
than 10Hz) for commercial devices.

This article presents one solution for accurate 27MHz clock recovery from MPEG-2 PCR
values with existing jitter during transport. It shows a unique way to design a clock recovery
loop with a VCXO built with a quartz crystal with a variable, varactor-based, capacitive
load. A closed-loop system is executing on a microcontroller, which in turn controls other
functions of the set-top box. It has been shown that a loop performance depends on
available memory space for additional buffering of demultiplexed, compressed bit streams.
Finally, a PLL and loop performance has been verified with NTSC color subcarrier
generation and measurement.

Branko Kovacevic received his BSEE and MSEE from the University of Belgrade in
Yugoslavia at the Faculty for Electrical Engineering in 1988 and 1994, respectively.
From September 1988 to June 1993 he worked in the R&D Institute "Mihailo Pupin"
in the Robotics and Mechatronics Department on a design of x86
multiprocessor-based robot controllers. As a design engineer, Branko worked on the
development of analog and MPEG-2 digital consumer satellite receivers for
multistream, point-to-multipoint, multicarrier direct broadcasting satellite (DBS)
networks. From the summer of 1997, Branko works for ATI Technologies, Inc. in
Toronto, Canada. He can be reached by email at bkovacev@atitech.ca.