Boonton 4240 RF Power Meter - Manual (Page 51)

Boonton 4240 Series RF Power Meter
4.4 Filtering
The Model 4240 employs digital filtering (averaging of measurements) to reduce the noise floor of the instrument and to
stabilize measurements. The default values are optimized for speed and low noise under general conditions. Default values
for normal and fast modes are as follows:
Range Normal (sec.) Fast (sec.)
0 2.8 2.8
1 0.8 0.8
2 0.8 0
3 0.8 0
4 0.8 0
5 0.8 0
6 0.8 0
The filtering technique used is digital pipeline filtering, also referred to as circular filtering or moving average filtering. The
displayed measurement is simply an equally weighted average of the last x seconds worth of samples, where x is the filter
length in seconds. For purposes of noise and settling time, the number of samples is not important, but the time is important.
For example, if a three second filter is used, the noise is the same whether 60 or 600 samples are taken in that interval,
provided that the samples are taken above a certain rate. For this reason, filter selection in the Model 4240 is done on the
basis of seconds, rather than the number of samples.
The bottom end sensitivity of the instrument is limited by sensor noise. An RMS noise specification is valid since the sensor
noise and the amplifier noise are band-limited and Gaussian. The noise level, specified in picowatts at a certain filter length,
is sufficient to calculate the error due to noise at any signal level, for any filter, as shown in the discussion of noise that
follows.
4.5 Noise
Noise Reduction. The amount of noise reduction that can be realized has no theoretical limitation, except that drift enters
into the picture at filter lengths over 20 seconds. The digital filter has a bandwidth and rolloff curve just as any filter does; the
bandwidth can be reduced arbitrarily. The effective noise bandwidth is 0.469/t, where it is the filter length. For example, with
a filter length of 4 seconds, the equivalent noise bandwidth is 0.12 Hz.
Figure 4-5 is a nomograph showing the noise reduction that applies for various filter lengths, given the sensor noise with 2.8
second filtering. (This is the time for which diode sensor noise is specified.) Noise power is inversely proportional to the
square root of the filter length. Normally, noise power varies directly with filter bandwidth; however, because power sensors
are square-law devices (detected voltage is proportional to power), the noise power is proportional to the square root of the
bandwidth. This can be demonstrated with noise measurements. At very low filter lengths (less than 150 milliseconds),
however, the noise does not increase without bound for all sensors because the input amplifier noise is restricted with
hardware filters. This additional filtering is not shown in the nomograph.
Error Computation. Since the noise is Gaussian, both before and after filtering, statistics show the level of confidence
factor that can be associated with a given reading. (At medium and high power levels, the confidence factor is essentially
unity.) Figure 4-6 shows a typical set of samples and a typical error band specification of 2 sigma. Under these conditions,
95.4% of the readings will fall within +2 sigma.
Figure 4-7 shows the confidence factor for other error bands. The error band is expressed in pW, regardless of the power
level. (The percentage error band can also be calculated as shown below.) The RMS noise is taken from the sensor
specifications and modified as necessary for filter lengths other than 2.8 seconds. Knowing any two of the three parameters
(error band, RMS noise, and confidence factor), the third can be computed. For example, if the sensor RMS noise is 65 pW
and the confidence factor is to be 95.4%, the
sided (+130 pW). If this were the case, at a
measurement level of 1300 pW the percent erro ding to about +0.44 dB.
error band is 130 pW, single
r band would be 10%, correspon
Operation
4-5