LTC6655/LTC6655LN
19
Rev. H
For more information www.analog.com
performance can be further improved by wiring several
LTC6655s in parallel as shown in the Typical Applications
section. With this technique the noise is reduced by √N,
where N is the number of LTC6655s in parallel.
Noise Specification
Noise in any frequency band is a random function based
on physical properties such as thermal noise, shot noise,
and flicker noise. The most precise way to specify a ran-
dom error such as noise is in terms of its statistics, for
example as an RMS value. This allows for relatively simple
maximum error estimation, generally involving assump-
tions about noise bandwidth and crest factor. Unlike wide-
band noise, low frequency noise, typically specified in
a 0.1Hz to 10Hz band, has traditionally been specified
in terms of expected error, illustrated as peak-to-peak
error. Low frequency noise is generally measured with
an oscilloscope over a 10 second time frame. This is a
pragmatic approach, given that it can be difficult to mea-
sure noise accurately at low frequencies, and that it can
also be difficult to agree on the statistical characteristics
of the noise, since flicker noise dominates the spectral
density. While practical, a random sampling of 10 second
intervals is an inadequate method for representation of
low frequency noise, especially for systems where this
noise is a dominant limit of system performance. Given
the random nature of noise, the output voltage may be
observed over many time intervals, each giving different
results. Noise specifications that were determined using
this method are prone to subjectivity, and will tend toward
a mean statistical value, rather than the maximum noise
that is likely to be produced by the device in question.
Because the majority of voltage reference data sheets
express low frequency noise as a typical number, and as it
tends to be illustrated with a repeatable plot near the mean
of a distribution of peak-to-peak values, the LTC6655 data
sheet provides a similarly defined typical specification in
order to allow a reasonable direct comparison against
similar products. Data produced with this method gener-
ally suggests that in a series of 10 second output voltage
measurements, at least half the observations should have
a peak-to-peak value that is below this number. For exam-
ple, the LTC6655-2.5 measures less than 0.25ppmP-P in
at least 50% of the 10 second observations.
APPLICATIONS INFORMATION
As mentioned above, the statistical distribution of noise
is such that if observed for long periods of time, the
peak error in output voltage due to noise may be much
larger than that observed in a smaller interval. The likely
maximum error due to noise is often estimated using
the RMS value, multiplied by an estimated crest factor,
assumed to be in the range of 6 to 8.4. This maximum
possible value will only be observed if the output voltage
is measured for very long periods of time. Therefore,
in addition to the common method, a more thorough
approach to measuring noise has been used for the
LTC6655 (described in detail in Analog Devices AN124)
that allows more information to be obtained from the
result. In particular, this method characterizes the noise
over a significantly greater length of time, resulting in a
more complete description of low frequency noise. The
peak-to-peak voltage is measured for 10 second inter-
vals over hundreds of intervals. In addition, an electronic
peak-detect circuit stores an objective value for each inter-
val. The results are then summarized in terms of the frac-
tion of measurement intervals for which observed noise
is below a specified level. For example, the LTC6655-2.5
measures less than 0.27ppmP-P in 80% of the measure-
ment intervals, and less than 0.295ppmP-P in 95% of
observation intervals. This statistical variation in noise
is illustrated in Table2 and Figure18. The test circuit is
shown in Figure17.
Table2.
Low Frequency Noise (ppmP-P)
50% 0.246
60% 0.252
70% 0.260
80% 0.268
90% 0.292
This method of testing low frequency noise is superior to
more common methods. The results yield a comprehen-
sive statistical description, rather than a single observa-
tion. In addition, the direct measurement of output voltage
over time gives an actual representation of peak noise,
rather than an estimate based on statistical assumptions
such as crest factor. Additional information can be derived
from a measurement of low frequency noise spectral den-
sity, as shown in Figure19.