Recently, a reader wrote to me seeking a copy of a paper on measuring the transient peak-to-average power ratio of DTV signals. So perhaps the topic of DTV signal power, which is so different from analog TV signal power, is of interest to many readers.
The power of an analog signal is always measured as peak power, which is the power level reached while transmitting the synchronizing pulses. The FCC rules are written about peak visual power because it is easily and accurately measured. The average power of an analog TV signal fluctuates widely with the brightness of the scene. The average power is inversely related to the scene brightness, or its DC level.
A dark scene has a very low average picture level, while a very bright scene has a high average picture level. Furthermore, the average scene in a film is somewhat different in brightness than the average scene of a program shot by a studio camera.
The average power of a DTV signal is independent of scene content, while the transient peak power is a statistical property of the datastream being transmitted. FCC rules require measurement of the average power of DTV signals. In theory, transient peaks exceed the average power by a very large ratio--6 to 10 dB if one waits long enough to capture such elusive events.
The 6 dB transient peak-to-average power ratio was measured by the Advanced Television Test Center in 1995 at the output of the 8-VSB exciter used in testing the Grand Alliance DTV System, now the North American DTV standard.
Specifically, the 6 dB value is not exceeded 99.9 percent of the time. Put another way, transient peaks greater than the average power plus 6 dB were found 0.071 percent of the time. Higher values have been reported, but are less frequent. These figures are for the DTV exciter output. I am not aware of what would be measured at the output of a DTV transmitter, but it would be lower than exciter output due to compression in the high-power stages of the transmitter.
The FCC does not require broadcasters to measure their transient peak-to-average power ratio. They are required to measure the average power output of the DTV transmitter.
For a broadcaster considering using his NTSC transmitter for DTV, this 6 dB transient peak-to-average power ratio is important. Suppose the NTSC transmitter is able to provide 5 MW of effective radiated power at peak of sync. This is 37 dB above 1 kW, or 37 dBK.
The transient peaks of our DTV signal must pass through the transmitter, and since those peaks are 6 dB above the average power of the DTV signal, one might argue that we could get up to 31 dBK ERP (average) with an NTSC transmitter setup.
The maximum ERP for DTV in the UHF band at 615 MHz is 30 dBK. However, some transmitters would not meet the linearity requirements implied by the FCC DTV RF mask without some power back-off.
SIGNAL BEHAVIOR
Transient peak-to-average power ratio refers to the characteristic of digital signals in a bandwidth-limited channel. The easiest way to think of this is to start with an analog baseband video signal. Before an analog baseband signal can be digitized, it must be restricted in its spectrum so the sampling process does not generate aliasing.
If the sampling rate is 13.5 MHz, all video frequency components more than half the sampling rate must be removed by a low-pass filter. So 6.75 MHz is perhaps 40 dB down. We might like to pass video frequencies up to, say, 6.25 MHz, but a filter that attenuates 6.75 MHz by 40 dB and is flat to 6.25 MHz will have very poor transient response due to its sharp cutoff in the frequency domain. That is, it would have large pre-shoot and overshoot. Without digressing too much, a phase linear filter will produce equal pre-shoot and overshoot.
These transients--pre-shoot and overshoot--are undesirable. They increase the voltage swing to be digitized. This applies also to a bandpass filter, which restricts the DTV signal to 5.38 MHz so it fits inside a 6 MHz channel. These pre-shoots and overshoots are fundamental to sharp cutoff filters. They are called Gibbs Phenomena after the mathematician who discovered them. They become part of the signal and cannot be removed, even by clipping.
The details of how we measured the transient peak-to-average power ratio of our DTV signal are included in a paper I wrote entitled "Measuring Peak and Average Power of Digitally Modulated Advanced Television Systems," published in the December 1992 issue of IEEE Transactions on Broadcasting. I have a few pre-prints available and will mail one upon request.
MEASUREMENT
Perhaps the most important part of that paper is a discussion of how to measure the average power of DTV signals. The average power is either the RMS voltage squared divided by the load impedance or the RMS current squared, multiplied by the load impedance. The classic method to measure RF power is to measure the heat produced, which is proportional to the RMS voltage or current.
Controlling all the parameters involved is tricky and time consuming. Moreover, this is an out-of-service measurement with the transmitter power being dissipated in a load resistor. However, it is a direct method and that's important, especially in testing transmitters at the factory or upon installation. Indirect in-service techniques sample the power output of the transmitter with a precision-directional coupler. The insertion loss of the directional coupler must be known to an accuracy better than the power measurement required to allow for calibration errors in the power measuring instrument.
This brings us to the power measuring instrument. There is the spectrum analyzer or vector spectrum analyzer, to which we will return, and there are thermal-sensing RF power meters and diode-based RF power measuring instruments.
Diode-based RF power meters are based on the fact that the square of the current through a forward-biased diode is proportional to the rectified RF power being measured. Now this is true over a certain range of currents for most diodes, but above this range, the current does not follow this square law. That means the diode-based instrument must be operated within its square law range of powers.
(click thumbnail)In the case of a complex signal like a DTV signal, the transient peaks must not be allowed to exceed the square law power limit of the instrument. If you do exceed the square law power range of such an instrument, it will read higher than the actual power.
Fig. 2 shows that a diode power meter was quite accurate for input powers below -20 dBm; that at -10 dBm, it read about 1.6 dB high, and at -5 dBm, it was 3 dB high, increasing to a 4 dB error at 0 dBm power input. Those measurements suggest that broadcasters should be very careful in the use of a diode power meter, because the possible errors might cause them to operate at a significantly lower ERP than is allowed.
Fig. 1 shows how we calibrated our diode power meter against a thermocouple-based RF power meter for unmodulated (sine wave) RF, DTV and white noise to produce the data reported in Fig. 2.
(click thumbnail)
White noise has a much higher transient peak power than the 32 QAM or 4-VSB modulated DTV signals discussed in my 1992 paper and shown in Fig. 2.
Thermal sensing RF power meters measure the heating effect of the sampled RF power, which is typically less than a milliwatt. Modern thermal-sensing RF power meters automatically cancel out the effect of the surrounding ambient temperature, but they have an inherent noise floor. The RF power sample should be at least 26 dB above the instrument's noise floor, or you will have to calculate the effect of the instrument noise.
These instruments should not be overloaded!
A spectrum analyzer or vector spectrum analyzer can also measure the average power of DTV signals indirectly (with a calibrated directional coupler). Here again, there are certain sources of error which must be avoided. The pilot carrier of our 8-VSB modulated DTV signal is an unmodulated spectral component, so it occupies zero bandwidth. The sidebands occupy the entire 6 MHz. The resolution bandwidth of the spectrum analyzer changes the displayed spectral power density of the sidebands over a large range relative to the displayed pilot carrier.
If the resolution bandwidth is 10 kHz, and the scan time long enough to get correct results, the indicated average power is quite low compared to the indicated power with say, 500 kHz resolution bandwidth. At 500 kHz, the indicated power is low compared to the RF sample being measured. The correction for resolution bandwidth (res BW) is 10log10 res BW/5.38 MHz, so for 500 kHz res BW, the indicated power is 10.3 dB below its actual value. For a res BW of 0.010 MHz, it is down 27.3 dB from its actual value. I cite these numbers to alert you to the enormous errors possible due to the effect of the resolution bandwidth selected by the operator.
The problems don't stop here. There are other correction factors having to do with the shape factor of the instrument's resolution bandwidth filter, and yet another having to do with the internal operations of the analyzer. In short, these sources or errors must be understood for each instrument. The instruction book for your spectrum analyzer can keep you out of trouble, read it carefully. These other error sources are smaller than the effect of resolution bandwidth, but they are still significant.
All indirect RF power measurements are subject to errors. You should fully understand how your RF power meter works and understand how to calculate the station ERP from these indirect measurements, a topic reserved for a later issue. My thanks for reader Chuck Condie for raising the subject of this paper from his request for my 1992 IEEE contribution.
