Model 1133A Power Sentinel TM Functional Description

Overview Voltage Inputs

® TM
The Arbiter Systems Model 1133A Power Sentinel The voltage inputs use low-TC voltage divider resistor
consists of several blocks. They are: GPS receiver and networks (attenuators). This input may be configured as
synchronization; voltage and current inputs; program- a three-phase, three-element input, with four connec-
mable-gain amplifiers, multiplexers, and analog-to-digi- tions (A, B, C, and optionally N); or it may be configured
tal converter; digital signal processor; host processor; as a two-element input, with independent connections to
display and keyboard; I/O functions; and power supply. each element (A+, A–, C+, and C–, for example).
The key features of the instrument and its measurement
functions are described in the sections which follow.
Self Calibration
The Model 1133A uses two processors to handle the
many tasks which are performed each second in such a The input sections are designed to provide exceptional
highly-integrated instrument. The digital signal analysis stability over time and temperature through the use of
tasks are performed by a TI TMS320C32 floating-point several high-performance (and fairly expensive) com-
DSP unit. The I/O functions and interface to the “real ponents. The rest of the analog circuitry is implemented
world” are handled by an ST 10F167 16-bit with lower-cost components. To reduce (essentially elimi-
microcontroller. Both of these processors have signifi- nate, in fact) the effects of drift and temperature sensi-
cant additional processing power which is not used in tivity in the programmable gain amplifiers (PGAs) and
the initial version of the Model 1133A. This will allow for analog-to-digital converter (ADC), an internal self-cali-
the future expansion of features, which is sure to come, bration source is provided. The inputs to the PGAs are
with a minimum amount of upset; indeed, many future multiplexed, each to four different signals: the three
enhancements should be possible with only a firmware current (or voltage) inputs, and the calibration signal.
upgrade.

Error Analysis
GPS Synchronization
All identified sources of error in the Model 1133A have
The Model 1133A includes an eight-channel global posi- been quantified using worst-case manufacturers’ per-
tioning system (GPS) receiver which provides time formance data. These have then been combined using a
accurate to a fraction of a microsecond anywhere in the root-sum-of-squares (RSS) method to yield a perfor-
world. Using proprietary technology developed by Arbi- mance estimate. Effects due to initial calibration, mea-
ter Systems and refined in several generations of GPS surement noise, temperature, and aging are all included.
timing products, an internal 10 MHz crystal oscillator is The reason for using RSS analysis is beyond the scope
slaved to the 1PPS output of the GPS receiver, maintain- of this paper. However, we have found in our many
10
ing its frequency at any time within a few parts in 10 . years’ experience building calibration instruments that
All of the internal timing signals are derived from this this method yields the most realistic estimate of actual
accurate timebase. worst-case performance, provided that numerous errors
contribute significantly to the overall performance (i.e.,
no one error dominates), and provided that actual worst-
Current Inputs case data is used in the analysis.

The Model 1133A has a three-phase current input which The error analysis for the Model 1133A for power or
is designed to be accurate to a few ppm over time and energy measurements is shown in table 1. Similar
temperature (see Error Analysis). This current input analyses have been performed for the other functions of
section uses a two-stage process, similar to a two-stage the instrument, and include most of the same factors
current transformer except that the first stage is dc- shown here.
coupled. Each of these two stages has an accuracy of a
few tenths of one percent; together, they have an
accuracy of a few ppm.

Arbiter Systems, Inc. · 1324 Vendels Circle, Ste 121 · Paso Robles, CA 93446
Tel +1 805 237 3831 · Fax +1 805 238 5717 · E-mail sales@arbiter.com · http://www.arbiter.com 1
Model 1133A Power Sentinel TM Functional Description

Table 1 – Power/Energy Error Analysis Power and Energy

Temperature Errors, 0-50° C: Error, ppm Power and energy are determined by making twenty
Current input, resistor TCR 63 separate measurements per second of the cross-prod-
uct of voltage and current for each phase. Each mea-
Voltage input, resistor ratio 50 surement uses 1024 samples (i.e., it takes data from a
100 millisecond window), yielding a 50% overlap.
Voltage reference (x2) 50
Apparent power (VA) and reactive power (VAR) are
Cal. source, resistor ratio (x2) 25 determined from the results of the voltage and current
Time Stability, 1 year: magnitude measurements (see next section), using stan-
dard identities. The active power measurements and
Current input, resistor 25 reactive power measurements are then compensated
for PT and CT corrections (see below) using a complex
Voltage input, resistor ratio <20 multiplication, and corrections are performed for trans-
former iron and copper losses, if enabled.
Voltage reference (x2) 36

Cal. source, resistor ratio (x2) <40 At this point, two different things are done with the
resulting measurement data. First, a determination is
Measurement Noise 10 made of the quadrant in which this data should be
registered (Wh delivered or received, VARh delivered or
Total RSS Error, Basic 117 received.) The results of each measurement cycle (20/
sec) are then added to the proper set of registers. These
Calibration Errors:
registers are stored periodically, and accumulation re-
Cal. artifact, Rotek MSB-001A 50 started from zero. The user may configure the unit for
different intervals to register energy.
Traceability to National Standards 50
The second thing that is done with the data is to deter-
Total Error, RSS 136 mine the actual power level. This number will be dis-
played on the front panel (as watts or VARs), and it will
Specification (0.025%) 250
be returned via the serial interface if a simple request for
“power” is made. This result is calculated by averaging
Signal Sampling the 20 power measurements made during each second.
Therefore, the update rate for this quantity is once per
second. This data is not registered separately depend-
The signals, representing the scaled voltage and current
ing on quadrant, as the energy data is; therefore, it is
inputs and the calibration signal, are time-multiplexed
theoretically possible, if the direction of power flow
into a two-channel ADC. Each signal is sampled at a rate
changes periodically, that the sum of measurements
of 10240 samples per second. The current and voltage
reported over the serial interface may gradually depart
signals for each phase are sampled simultaneously to
from the registered energy data. This is due to the loss
eliminate errors in power calculation which would occur
of information in the averaging process; the registered
with non-simultaneous sampling. The sample clock in
data is the most accurate. The averaged data is provided
the Model 1133A is synchronized to UTC-USNO (GPS)
primarily as a convenience or for system control pur-
within one microsecond, allowing measurements of phase
poses; it is not intended for billing purposes.
angle across a power grid to be compared directly, and
ensuring that revenue is billed at the correct rates.
Voltage and Current

Voltage and current are measured in a similar fashion to

power, using overlapping 1024-point measurements. In
this case, however, the cross product is replaced with

Arbiter Systems, Inc. · 1324 Vendels Circle, Ste 121 · Paso Robles, CA 93446
2 Tel +1 805 237 3831 · Fax +1 805 238 5717 · E-mail sales@arbiter.com · http://www.arbiter.com
 Model 1133A Power Sentinel TM Functional Description

2
the square of the voltage or current samples. The square Copper compensation is used to correct for the I R
root of the resulting sum is proportional to the rms voltage losses in the transformer windings due primarily to their
or current value during the measurement interval. This (non-zero) resistance. As you would expect, this effect
value is corrected for the CT and/or PT correction is primarily active (resistive), although there may be
factors before further use. minor reactive effects, and it is proportional to the
current squared. This factor allows the user to correct
The resulting data is used to correct the energy mea- for these losses. It is a complex factor, providing both
surements, as described above, and averaged over a watts and VARs correction, and is proportional to cur-
one-second interval to provide data for display and rent squared; i.e. so many watts and VARs are to be
reporting to a host system. subtracted from the registered amounts per ampere
squared of load current.

CT and PT Compensation Iron losses (also called core losses) are due to magne-
tizing currents (the small amount of current required to
CT and PT compensation may be enabled to correct for generate the flux in the core, which is unrelated to the
the inaccuracies of the CTs and PTs used in the load current) and eddy current losses in the core
metering setup. The system voltage is relatively con- material. These are approximately proportional to the
stant, so the PT compensation factor is a single, complex square of the voltage, and the compensation is per-
(i.e., magnitude and phase, or real and imaginary) formed using the same basic method as described
correction factor. above for copper loss.

CT compensation is more complicated. Due to the fact

that magnetizing currents in CTs are not exactly propor- DC Offsets
tional to the load current, a matrix is used, allowing the
entry of several different compensation factors mea- DC offsets may be present in the signals applied to the
sured at different current levels. The Model 1133A input of the Model 1133A, although this is unusual. More
interpolates between the numbers in this table (also commonly, small dc errors in the measurement circuits
complex) to determine the correction factor to be used. of the unit itself result in non-zero average of the
samples.
Correction for energy is performed using the (complex)
product of the PT and CT factors. Correction for voltage Since the Model 1133A makes wideband measurements
or current is performed using the magnitude of the of power, voltage, and current (i.e., components at any
appropriate factor. Correction for phase angle is made frequency within the measurement bandwidth – which
using the phase of the appropriate component, i.e. the includes dc – will affect the measurement), this is a
arctangent of the complex value. The actual calculation potential source of error which must be corrected to
performed may be different than this description, due to obtain maximum accuracy. Therefore, part of the mea-
computational considerations (a complex multiplication surement process is to average the (windowed) data,
is far faster than a trigonometric operation such as an measuring the dc component. The effects of this compo-
arctangent, for example); however, the end result will be nent are then subtracted from the results.
as described.

Phase and Frequency

Transformer Compensation
As a part of its measurement process, the Model 1133A
There are two different types of transformer compensa- performs a fast Fourier transform (FFT) of the windowed
tion. They are used to correct for the losses in a voltage and current samples. In accordance with IEC
transformer when primary-side metering is used to 1000-4-7, this process is performed twenty times per
meter the energy delivered to a customer at the second- second, using overlapping 1024-sample hanning-win-
ary of the transformer. dowed data. This yields new FFT results twenty times per

Arbiter Systems, Inc. · 1324 Vendels Circle, Ste 121 · Paso Robles, CA 93446
Tel +1 805 237 3831 · Fax +1 805 238 5717 · E-mail sales@arbiter.com · http://www.arbiter.com 3
Model 1133A Power Sentinel TM Functional Description

second for each voltage and current input, for a total of There is an error in this approximation, due to the fact
120 FFTs per second. Phase angle may be determined that there will be a small amount of energy leakage into
from the relationship between the real and imaginary nearby bins, which will not be included in the three
component of the fundamental-frequency bin of the which are measured. This is generally of little conse-
FFT. (Since the window is 100 ms wide, each bin is 10 quence, as it turns out. First of all, when the frequency
Hz apart; therefore, this is bin 5 for 50 Hz and bin 6 for is close to accurate (which is most of the time), then the
60 Hz.) amount of energy outside of the three bins summed is
very low. As a matter of fact, with the hanning window,
So long as there is significant measured energy in the there is no signal at all outside of these three bins if the
bin, frequency offsets do not affect the measured phase frequency is exact.
angle. This is true as long as the signal being measured
is the main source of energy in the bin, i.e. there is In the real world, however, the frequency will be off
minimal leakage from adjacent bins, and minimal noise. somewhat, and it is reasonable to ask how large the
Provided that the frequency is anywhere near nominal error can be. For small frequency errors, say 0.01 Hz,
th
(within 10 Hz or so), the phase measurement is perfectly the 50 harmonic will be 0.5 Hz from the center of the
usable. nominal bin. This results in an error of about 0.005% –
insignificant. The worst-case error will occur when a
The phase measurements may be compared to deter- harmonic is very nearly centered between two bins. In
mine phase angle between voltages and currents, or this case, the algorithm described above will ‘miss’ a bin
between any two voltages or currents. Because the containing a signal with an amplitude of about 17% of
sampling process is synchronized via GPS to UTC, the actual harmonic amplitude. The energy contained in
2
absolute phase angle measurements may be made and this bin is then (0.17) or about 2.9% of the total energy,
compared between two units located at some distance resulting in a measured energy 0.971x what it should be.
0.5
from one another. The measured harmonic amplitude will then be (0.971)
or about –1.5% in error. This is well within the specifica-
Frequency is measured by taking the difference in tion limits (5%) of IEC 1000-4-7. This worst-case error
th
phase angle between subsequent measurements, based would occur for the 50 harmonic with a fundamental
on the identity f = dØ/dt. Frequency is averaged over frequency error of 0.1Hz. At lower harmonics, the
one second prior to being displayed or made available frequency error must be progressively greater; for
for output. example, to result in a –1.5% error in measuring the 9th
harmonic would require a fundamental frequency offset
of 0.556 Hz.
Harmonics
The phase angle of the harmonics can be determined by
Harmonics are measured using overlapping hanning- taking the arctangent of the real and imaginary compo-
windowed FFTs of 1024 samples and 100 ms window nents of the bin closest to each harmonic. This informa-
length. Based on the instantaneous frequency, the loca- tion cannot be used, however, in the averaging process
tion of the bins containing significant energy for each described in IEC 1000-4-7, because this specification
harmonic are determined. This is a total of three bins, requires the rms average of a series of measurements.
one approximately centered on the harmonic and those This, by definition, requires magnitude data only. There-
two adjacent to it. Then, the energy in those three bins is fore, the harmonic phase is not normally calculated, but
totaled, resulting in the energy for that harmonic. This it can be requested or displayed. In this case, the result
can then easily be expressed as a percent of the rms shown will be the phase angle determined by a single
signal level, or in whatever form is required. While there measurement, and it will be calculated ‘on-demand,’ i.e.
is a closed-form correction which can be employed to when it is requested only.
find the harmonic magnitude in the presence of fre-
quency errors, this approximation was chosen because
it is much faster and gives adequate performance.

Arbiter Systems, Inc. · 1324 Vendels Circle, Ste 121 · Paso Robles, CA 93446
4 Tel +1 805 237 3831 · Fax +1 805 238 5717 · E-mail sales@arbiter.com · http://www.arbiter.com
 Model 1133A Power Sentinel TM Functional Description

Flicker System Time Deviation

Flicker is measured in accordance with IEC 1000-4-15, System time deviation, which is the accumulated error of
the successor standard to IEC 868. Unlike the other a clock using the system frequency as its reference,
measurements described above, flicker measurement is compared to an absolute reference such as UTC, is
a continuous process. This process is performed using determined from the 20/second phase data described
a sample rate of 640 samples per second (sps). Anti- earlier. System time deviation is accumulated as integer
alias (decimation) filtering is performed on the 10240 cycles of error plus fractional phase, and is converted to
sps data stream, and the resulting samples are further seconds as needed. The A-phase voltage phase angle is
processed following the block diagram suggested in IEC used for this measurement. Since this is an integrated
1000-4-15. The resulting measurements of flicker per- value, the constant of integration (initial time offset value)
ceptibility are classified using a 256-level logarithmic must be specified by the user.
classifier at the full 640 sps rate. Pst is then determined
each ten minutes as described by the standards.
Phasor Measurements
Although no standards currently require it, the Model
1133A also measures flicker on the current inputs. This Phasor data are formatted and output in accordance
information can be useful in determining whether a with IEEE Standard 1344-1995. Phasors consist of the
customer’s load is causing flicker on the power system, real and imaginary component of magnitude for the
or whether the customer is being subjected to flicker voltages and currents at a particular point in a power
from other sources. It is unrealistic to penalize a utility distribution system, along with suitable time synchroni-
for ‘poor power quality’ at a customer’s load when the zation fields and other information. This information is
cause of the problem is the load itself. available in real time, and is based on the measured
fundamental voltage, current, and phase angle described
above, at a 20 records per second rate. There is a
Interruptions measurement delay due to the data acquisition delay of
50 ms, signal processing time of approximately 15 ms,
Interruptions are monitored on the voltage inputs by and data transfer time which depends on the data rate.
comparing the 20/second voltage measurements with a
user-supplied threshold. Events where the voltage dips
below the threshold trigger an event, which can cause Phase Balance
the logging of pre- and post-fault data, contact closure,
or any of the other actions described under ‘Event The Model 1133A measures phase balance by calculat-
Logging.’ ing the symmetrical sequence components (positive,
negative, and zero sequence) for the three-phase volt-
age input. Normally, if the unit is connected properly, the
Voltage Fluctuations positive-sequence voltage will be equal to the line voltage
and the negative- and zero-sequence voltages will be
Voltage fluctuations are monitored by classifying the 20/ approximately zero. In the event that a user-specified
second voltage data, per phase, with a 256-step linear limit on the imbalance (as a percentage of nominal
classifier covering a range of ±20% of nominal voltage. voltage) is exceeded, an event will be recognized. In
These data are then summarized as a cumulative prob- addition, the sequence components are averaged over a
ability table over a specified interval, typically 15 min- user-specified interval (typically 10 or 15 minutes) and
utes. In addition, the minimum, maximum, mean and may be logged if desired. These calculations are per-
standard deviation are calculated. The data may be formed using the voltage magnitude and phase informa-
recorded in flash memory either continuously or on tion, 20/second.
demand.

Arbiter Systems, Inc. · 1324 Vendels Circle, Ste 121 · Paso Robles, CA 93446
Tel +1 805 237 3831 · Fax +1 805 238 5717 · E-mail sales@arbiter.com · http://www.arbiter.com 5
Model 1133A Power Sentinel TM Functional Description

Load Balance and type of events cannot be known a priori, the amount
of memory required also cannot be stated with certainty.
Load balance is calculated in much the same way as The Model 1133A handles this by allocating sufficient
phase balance, except on the customer’s load current. memory as required for the registered quantities over
Both of these measurements may be used to identify the specified period of time, and then making whatever
serious power system problems, such as a dropped memory is left available for event logging. You can
phase, which could cause serious damage to both the configure the Model 1133A to initiate an auto-dial call or
utility’s and the customer’s equipment. contact closure when the remaining event memory (or
register memory) drops below a specified limit. Memory
is allocated in 64Kbyte blocks, which is the block size of
Flash Memory and Event Logging the flash memory.

Flash memory must be erased in blocks, and data

The Model 1133A includes 2 megabytes standard of
cannot be over-written until its block is completely erased.
flash memory for data and event logging. Up to 8
Therefore, the normal process will be to first read out the
megabytes total may be optionally installed. This memory
desired data, and then erase the blocks, making them
may be used to record two basic types of information:
available for re-use. Each block of data is password-
registered quantities, which are recorded on a fixed
protected, having two levels of security: one to access
schedule (and may include many more items than a
the data, and a second to clear it from memory. The
typical energy meter, including just about any function
registered data may be separated into two blocks, each
the instrument can measure); and event data. Event data
with its own security. This would normally be used to
is logged on the occurrence of an event designated to
separate billing data from operational data.
initiate logging.
In addition to the password security, the flash memory is
These events may consist of a measured quantity ex-
located on a separate module, mounted internally to the
ceeding a user-specified threshold, an external trigger,
Model 1133A. In the event of a failure or sabotage to the
or an internal state of the Model 1133A (low flash
instrument itself, the memory module may be removed
memory, for example). In addition to causing the logging
and read on a separate Model 1133A mainframe, pre-
of specified data, an event may trigger a dial-up modem
venting the loss of critical billing information and other
call to be initiated, reporting the event; and/or a contact
data.
may be closed (or opened).

There is a great deal of flexibility as to what may be

Contact Outputs
recorded in flash memory at the time an event is recog-
nized. The time, type of event, state of any number of
measured quantities at the time of the measurement, Four contact outputs may be used to report events
pre-and post-event data, which may be measured quan- recognized by the Model 1133A, or they may be con-
tities (voltage, power, etc.) or waveform data, may all be trolled remotely, by command. They may also be oper-
recorded. The Model 1133A must be configured in ated on a schedule, which may be downloaded for up to
advance to specify the events being recognized and 30 days in advance. In addition to reporting events,
actions to be taken. these contacts may be used to synchronize external
equipment or to operate load-control switchgear.
There is a fundamental difference between registered
quantities and event log data. The amount of memory
needed to record a certain number of registered items Event Inputs
for a certain period can be determined exactly, whereas
the amount of memory required to log events depends Four external event inputs are also provided. These are
on the number and type of events, and the number of optically-isolated and accept dc signals at levels of 24 to
items to be recorded for each type. Since the number 240 volts. Upon the application of a signal to one of these

Arbiter Systems, Inc. · 1324 Vendels Circle, Ste 121 · Paso Robles, CA 93446
6 Tel +1 805 237 3831 · Fax +1 805 238 5717 · E-mail sales@arbiter.com · http://www.arbiter.com
 Model 1133A Power Sentinel TM Functional Description

inputs, it will be time-tagged to one microsecond resolu-

tion and recognized as an event, which can cause any
of the actions described above under ‘Event Logging.’

Serial Channels

The Model 1133A includes two serial channels. Each of

these can be configured at the time of order with RS-232
or RS-485 drivers, or with a V.34bis 33.6k modem. Each
channel can have access to all functions of the Model
1133A; alternatively, certain functions (such as the
ability to clear the revenue registers) may be enabled or
disabled independently. Each channel has an RJ11 (6-
position) modular connector with 2 meter (7 ft.) modular
interconnect cable , and the RS-232 and RS-485 func-
tions are provided with an adapter to DB-9 (male) plug.
A DB-9 female adapter, and longer cables, can be
supplied on special order. These adapters may be
rewired in the field to match interface requirements.

Both serial channels, and the Ethernet channel, may be

operated simultaneously, each serving different hosts
with whatever information each needs and is authorized
to access.

Ethernet MMS Interface

The Model 1133A contains a provision for an Ethernet

interface supporting the emerging MMS standard for
substation automation. This interface will have a 10BaseT
interface via an RJ45 (8-position) modular connector.
This interface will not be available initially, but the
hardware is included, to allow future expandability when
this capability is added. Contact the factory for more
information on this feature.

Arbiter Systems, Inc. · 1324 Vendels Circle, Ste 121 · Paso Robles, CA 93446
Tel +1 805 237 3831 · Fax +1 805 238 5717 · E-mail sales@arbiter.com · http://www.arbiter.com 7