SPE 84481

GT
E
Correction for Processing Multigradient, Multiple-T
E
NMR Log Data
S. Chen, SPE, G. Hursan, SPE, D. Beard, and D. Georgi, SPE, Baker Atlas, Houston, Texas, USA
Copyright 2003, Society of Petroleum Engineers Inc.

This paper was prepared for presentation at the SPE Annual Technical Conference and
Exhibition held in Denver, Colorado, U.S.A., 5 8 October 2003.

This paper was selected for presentation by an SPE Program Committee following review of
information contained in an abstract submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
for commercial purposes without the written consent of the Society of Petroleum Engineers is
prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300
words; illustrations may not be copied. The abstract must contain conspicuous
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.

Abstract
We have developed a data processing method that allows a
valid integration, in the time domain, of echo trains acquired
in different field gradients, G, and different interecho times,
T
E
. The combined echo train can be used to obtain clay bound
water (CBW), bulk volume irreducible (BVI), and total
porosity information with improved vertical resolution. The
same data, in the uncombined form, are used for fluid property
estimation. Thus, the data are used more economically.
As an added benefit, the combined echo train allows the
use of a single T
2cutoff
, consistent with the laboratory core-
NMR derived T
2cutoff
. Although using a zero-gradient,
laboratory core-NMR derived T
2cutoff
to calibrate NMR logs
acquired in a gradient field is a common practice, the
discrepancy cannot be ignored for NMR logs acquired with a
large GT
E
product. With the new method, this discrepancy is
minimized. Field examples are provided to show the benefit
with the new method.

Introduction
The new generation, multi-frequency NMR logging
instruments, such as the Magnetic Resonance Explorer
TM

(MREX) from Baker Atlas, are capable of acquiring data
useful for characterizing both the formation rock properties
(e.g., bound and movable fluids, porosity, and permeability)
and reservoir fluid properties. However, these different
properties often require an assortment of NMR acquisition
parameters and sequences. High-resolution formation rock
characteristics require acquisition schemes that generate
repetitive echo trains to reduce the vertical stacking
requirement. In contrast, fluid properties usually vary more
slowly with depth than rock properties and require acquisition
methods that can maximize the fluid contrasts, achievable with
variable magnetic field gradients (G), interecho times (T
E
),
and wait times (T
W
) in the acquisition scheme. Currently,
multiple G, T
E
, and T
W
data are not combined in the time
domain to obtain formation rock properties because echo
trains having different GT
E
cannot be simply stacked.
Common practice is to log either multiple passes, each for a
separate objective or a single, slow, comprehensive acquisition
pass. Either way the data are not used economically.
Baker Atlas global NMR logging service records reveal
that a high percentage of NMR logging job requests are for
formation rock property characterization. These properties
include total and effective porosities, clay and capillary bound
water volumes, movable fluid volume, and permeability. Light
hydrocarbon typing and flushed-zone gas or oil saturations
comprise the next largest group of logging job requests. There
are also less commonly exploited techniques of NMR logging,
such as those used in heavy oil formations, which have
experienced some success in various geological areas
worldwide. This global picture roughly reflects the relative
robustness of the NMR logging techniques available to-date.
Based on this assessment, we have developed a small-number
of objective-oriented acquisition sequences that coincides with
this global picture.
1
These single-pass, comprehensive
sequences all include an objective of formation rock property
estimation.
Ideally, the multifrequency, multiple echo trains generated
from a single acquisition pass are processed using a
comprehensive inversion method with a forward model based
on the knowledge of fluid and rock properties, thereby
providing all petrophysical information in a single processing
step.
1,2
Such an approach, however, is more suitable for
postprocessing because it may require interpretation input, for
example, known fluid properties. For well-site deliverables,
often the rock property information is sufficient for well
completion decisions. An automated, rapid processing method
for obtaining rock properties, requiring no interpretation input
from the log engineer, is essential. One frequently used
approach is to invert individual echo trains separately. The
porosity is derived from just one of the inversion results. Thus,
a great amount of data is not used. Another approach is to
stack all echo trains together that are acquired with the same
T
E
and are fully polarized, even though they are not acquired
at the same frequency and field gradient. The second approach
is valid only if the frequency and the corresponding field
gradient differences do not affect the overall decay rate
substantially. Otherwise, a smeared T
2
spectrum is obtained,
which, in some cases, can affect the BVI interpretation. We
propose a method to correct for the effect using a time-
dependent weighting function to stack all pertinent echo trains
E
T G
2 SPE 84481
together, thereby improving the data quality and reducing the
requirement for vertical stacking.

Features of the MREX Tool
Baker Atlas MREX tool possesses several important features.
Figure 1 illustrates the MREX tools static and RF field
distributions for one representative cross section. The tool has
a vertical aperture of 24 in. Extended magnets on both ends of
the antenna are used to provide prepolarization of the nuclei
under investigation. The details of the tool will be published in
a subsequent paper. In this paper, a brief summary of the tool
features is provided.
The MREX is a side-looking tool with a sensitive volume
that penetrates into the formation from 2.5 in. to 4.5 in.
depending on the operating frequency. This penetration depth
is selected to be tolerant of borehole irregularities and less
affected by invasion. The side-looking configuration does not
sense the borehole signal from behind the tool (Figure 1 and 2)
and, thus, it can be used for almost all borehole sizes.
Furthermore, centralization is not an issue for the MREX in
deviated and horizontal wells.
The side-looking MREX is a gradient tool. The gradient
strength decreases with frequency. The gradient feature makes
it possible to excite multiple sensitive volumes in one single
polarization time using a frequency interleaving method. The
operating frequency band ranges from mid-400kHz to upper-
800kHz with sufficient frequency separation between
frequencies to avoid interference from the excitations in
neighboring sensitive volumes. Because the magnitude of the
MREX tools field gradient varies with frequency, the echo
trains acquired with different frequencies may exhibit different
apparent decay, even if they are acquired with identical
acquisition parameters, such as interecho time, T
E
, and wait
time, T
W
. The longer the echo-train length is, the greater the
difference in apparent T
2
decay.


Figure 1. Illustration of one cross section of the B
0
and B
1
field
distribution from the MREX tool. The arrows in the squares
indicate the B
0
and B
1
directions in the center of the sensitive
volume range.

Multiple Echo Train Acquisition by MREX
The multi-frequency capability of the MREX tool allows us to
acquire a large number of echo trains in a single pass, yielding
both formation rock and fluid properties. For example, a six-
frequency version of the PoroPerm package and a three-
frequency PoroPerm + Oil package each acquire more than ten
echo trains. These echo trains all satisfy the following
expression, but the parameters are different:

|
|
.
|

\
|

|
|
.
|

\
|

=
comp N
i
fluid E j
E
i
E
i i
E j
D T G
kT
T
T k
M p
T k f E
m
m
m
_ 2 2 2
2
12
exp exp
) , , (
(1)

where i, j, k, m are indices for i
th
T
2
component, j
th
frequency,
k
th
echo, and m
th
T
E
, respectively, and

( )
i W i
T T p
l
1
exp 1 = (2)

is the polarization factor for the i
th
T
1
component and l
th
wait
time, T . The spread of frequencies, T
l
W
E
, T
W
, and the number
of echoes, N
E
, among these echo trains make their apparent
decay times significantly different. In this section, we describe
these two acquisition packages and their purposes. In the next
section, we describe a method for the time-domain integration
of these echo trains.

MREX MREX MREX
borehole
formation
B
1

B
0


Figure 2. Illustration of the side-looking MREX tools sensitive
volume (the red-colored area). Signals are contributed only from
the front of the tool over an aperture extending approximately
120.
Fiberglass
Sleeve

The PoroPerm package is designed to obtain basic
formation rock properties, such as effective and total porosity,
permeability, clay-bound and capillary-bound water volumes,
movable fluid volume, and the T
2
distribution.
Antenna
The left side of is a three-frequency version of the
PoroPerm acquisition package (PoroPerm3). The basic
acquisition scheme includes 7 echo trains (ET), which are
acquired with the same T
E
, typically 0.6 ms, but vary in
frequency, f, echo train length, N
E
, and T
W
. ET #1 is a fully
polarized, long echo train, typically with 500 ms of data. ETs
#2 and #7 are each a series of echo trainlets (ETL). Each has a
10-ms length (
E E
T N L = ) and a 30 ms wait time. During a 1s
acquisition window, 12 phase-alternating pairs of ETLs are
acquired and these ETL data are stacked in the acquisition
software. ETs #3 and #4 are fully polarized echo trains but
Figure 3
SPE 84481 3
with a data acquisition length of only 30 ms. These two ETs
can be combined with ET #1 to reduce the random noise in the
initial decay of the ET, which is critical to BVI estimation.
ETs #5 and #6 are partially polarized echo trains having the
same NE as #3 and #4. Combining ETs #1, 3, 4, 5, and 6
provides an alternative approach for BVI estimation.
Normally, ET #1 is acquired with the highest frequency and,
thus, its sensitive volume will be the closest to the borehole. In
the case of severe borehole rugosity, causing contamination in
the sensitive volume corresponding to the highest frequency,
ETs #3 and #4 will still provide the uncontaminated total
porosity, and the combined ET #3, 4, 5, and 6 will provide the
BVI estimate.


Figure 3. Illustration of PoroPerm acquisition package.

The three-frequency version of the PoroPerm acquisition is
used only when the resulting wait times for ETs #1, #3, and
#4 are sufficient to reach the complete polarization of all
signals. If the formation requires longer wait times for full
polarization, the six-frequency version (PoroPerm6), which is
shown on the right-hand side of Fig. 4, is used. We see that
PoroPerm6 simply repeats PoroPerm3; thus, the data
processing does not become more complicated. This approach
can be theoretically extended to 3n frequencies where n is an
integer; however, six frequencies usually suffice. Clearly, with
this implementation the number of echo trains acquired in a
unit time is independent of the number of frequencies.
There are advantages of implementing the variable
frequency number version instead of using a fixed, maximum
number of frequencies. First, the SNR is frequency dependent:
variable frequency implementation, we take advantage of the
higher frequencies and use the minimum number of
frequencies that can satisfy the need for continuous data
acquisition, while achieving full polarization. Second, when
fewer frequencies are used, the time between the two
consecutive phase-alternating pairs (PAPs) is reduced. This
results in a more flexible acquisition and delivers the data with
better vertical resolution.

A
d
j
u
s
t
a
b
l
e

T
i
m
e

g
a
p

3
A
3
C
3
E
3
F
3
D
1
A
1
B
2
A
2
C
2
B
1
C
1
C
2
C
T
W
L
T
W
L
T
W
L
T
W
S
T
W
S
A
d
j
u
s
t
a
b
l
e

T
i
m
e

g
a
p

3
A
3
C
3
E
3
F
3
D
A
d
j
u
s
t
a
b
l
e

T
i
m
e

g
a
p

3
A
3
C
3
E
3
F
3
D
1
A
1
B
2
A
2
C
2
B
1
C
1
C
2
C
T
W
L
T
W
L
T
W
L
T
W
S
T
W
S
F1 F2 F3

2
3
4
5
6

1
t
i
m
e
F
1
F
2
F
3

1
3
4
5
6

2
3
4
5
6

1
t
i
m
e
F
1
F
2
F
3
F
4
F
5
F
6

1
1
2
1
2
7 7
7
2 9
1
8
10
11
12
13
14

2
3
4
5
6

2
3
4
5
6

2
3
4
5
6

1
t
i
m
e
F
1
F
2
F
3

1
3
4
5
6

1
3
4
5
6

1
3
4
5
6

2
3
4
5
6

2
3
4
5
6

2
3
4
5
6

1
t
i
m
e
F
1
F
2
F
3
F
4
F
5
F
6

1
1
2
1
2
7 7
7
2 9
1
8
10
11
12
13
14
























the higher the frequency, the higher the SNR. By using a
igure 4. Three-frequency version of PoroPerm + Oil package.
The PoroPerm + OIL package (Figure 4) is designed for
log
trains. Table 1
sum
er usually requires adequate
polarization contrast and/or diffusion contrast. In the

F

n, a single-p
from fre
ging oil-bearing formations. Agai ass concept is
used with sufficient variations of T
E
, T
W
, and field gradient to
evaluate the formation and to perform hydrocarbon typing.
The acquisition is designed for wells having light and medium
grade oils with low to moderately high GOR.
The acquisition package yields 13 echo
marizes the default values of the key acquisition
parameters for each ET with the echo train IDs defined in
Figure 4. There are two pairs of dual-T
W
ETs (1A and 1B
quency #1 and 2A and 2B from frequency #2). The
implementation uses the highest frequency for frequency 1 (f
1
)
which couples with the longest T
E
(2.1 ms) for acquiring ET
1A and 1B. At this frequency, the tool gradient is
approximately 34 gauss/cm.
Discriminating oil and wat
4 SPE 84481
Por Effect of Multiple GT
E
on T
2cutoff
and BVI oPerm+OIL package, polarization contrast can be obtained
with two long echo trains having different T
W
s and can also
be obtained from a multi-component T
1
analysis. Diffusion
contrast is obtained using multiple
E
T G data. The maximum
E
T G is chosen to make the diffusion-induced decay,
( )
( )
2 2
D T G
D
E

=

, (3)
One of the important petrophysical parameters that can be
derived from NMR logs is the irreducible bound water
volume, BVI. A commonly used method for estimating BVI is
to use a T
2cutoff
such that


12
1
2diff
T

d laxation decay of the wate

substantially over the operating frequency range, we can
choose to use the highest frequency (thus, the highest
gradient) to match the longest T
E
, thereby maximizing the
diffusion contrast.
Alternatively, maximizing the
E
T G contrast results in a
great difference in t

=
cutoff
T
T
dT T P BVI
2
min 2
2 2
) ( (4)
ominate the overall re r phase.
Because the MREX tools magnetic field gradient varies
he observed ong individual echo
trai

where P(T
2
) is the apparent T
2
distribution with individual T
2

components given by:

1
diff 2
1
2
1
2

+ + = T
V
S
T T
B
. (5)
decay am
di ns. From Table 1 we see that three fferent T
E
s are used to
acquire the full-length (default value: ms 500
1
= =
i i
E E
N T L ),
fully-polarized echo trains, 1A, 2A, and 3A. Their
corresponding GT values are approxim
| | | | | | . 2 : 1 : 4 : :
3 2 1

E E E
T G T G T G


The extra decay due to diffusion, , expressed in Eq.
(3), depends on (a) the field gradient, G, which is associated
with the acquisition frequency, (b) the inter echo time, T
1
diff 2

T
E
, an
acquisition parameter, and (c) the diffusivity, D, a fluid
property. Obviously, data that is acquired with a different GT
E

combination results in a different apparent T
2
distribution,
P(T
2
). If the same T
2cutoff
value is used to compute the BVI of
the same formation, the results are gradient and T
E
dependent.
If the T
2diff
is significantly larger than both T
2B
and ( ) S V , the
dependency may be negligible. However, if the GT
E
product
is large, T becomes dominant and the diffusion effect may
not be negligible.
1
diff 2

E
ately
These differences must be corrected for be re the echo trains
an be integrated in the time-domain.
equency PoroPerm + OIL
package
(kHz) (ms) (s) (ms) group
fo
c

Table 1. Default parameters for the 3-fr
ET
ID
Freq TE TW NETE NS/
1A 880 2.1 4+* 500 1
1B 880 2.1 1 500 1
1C 880 0.6 0.03 10 12
2A 790 0.6 4 + 500 1
2B 790 0.6 1 500 1
2C 790 0.6 0.03 10 12
3A 695 1.5 4 + 500 1
3C 695 0.6 0.03 10 12
3D 695 0.6 0.05 10 8
3E 695 0.6 0.1 30 4
3F 695 0.6 0.2 50 4
: Acqu ead ti lus ad le d
Without considering the tool and acquisition dependencies,
the value of the T
2cutoff
depends on the rocks surface
mineralogy. Therefore, a common practice is to calibrate the
T
2cutoff
with laboratory NMR measurements. These lab
measurements are often performed in a uniform magnetic field
without an external gradient and, thus, a negligible
contribution from . To calibrate NMR logs using a
laboratory core-NMR derived T
1
diff 2

T
2cutoff
, one sometimes has to
consider the disparity of the field gradient between NMR
laboratory and logging data.
Figure 5 shows the gradient effect on the T
2cutoff
shift with
and without an external field gradient. All plots assume that
the BVI contains water only, at 200-F temperature
environment. The gradient strength is approximately 18
Gauss/cm for MRIL operating at 750 kHz and 26 Gauss/cm
for MREX at the same frequency. The intercepts of the
horizontal lines with the T
2cutoff
curves represent the apparent
T
2cutoff
values that should be used on the log data. Clearly,
there will be a dependence for carbonates if the GT
E
is large.
On the other hand, for a small GT
E
, a slight shift in the T
2cutoff

may not result in a significant BVI discrepancy.
* isition overh me p justab elay.

ffective
orosity, together with the ETLs for CBW, the total porosity
can
additional field-gradient induced decay differences.
quisition
ackages and log examples are described in Ref. 3.
The three fully polarized ETs are used for e
p
be obtained. There are five sets of short echo trainlets with
a length of 10 ms acquired with three frequencies (2 of 1C, 2
of 2C, and 1 of 3C), totaling 60 trainlets per sample. These
short, partially polarized trainlets are stacked together for
CBW estimation without the
E
T G correction because
their
E E
T N only equals 10ms, resulting in negligible

More details of PoroPerm and Properm+Oil ac
Although it is possible to adjust the T
2cutoff
according to the
gradient and T
E
, in practice it is confusing to use multiple
T
2cutoff
values for the same formation. Even if a T
2cutoff

adjustment is applied, individual echo trains still must be
inverted independently, missing the opportunity of combining
the multiple frequency data to obtain one single, high-quality
echo train.
p

SPE 84481 5


Figure 5. The discrepancy between T
2cutoff
values derived in the lab
with no applied gradient, and the apparent T
2cutoff
in the gradient
tool environment. All data are computed based on a 200-F
temperature environment.

Justification for GT
E
Correction
In this section we describe a method to reasonably integrate
echo trains acquired with different GT
E
combinations to
improve the BVI and total porosity estimation. The movable
fluid volume can be estimated once the porosity, BVI, and
CBW are known:

BVI CBW BVM
T
= . (6)

Thus, as long as CBW and BVI can be estimated accurately, it
is not essential to have an accurate T
2
distribution to obtain an
accurate BVM.
Generally, BVI contains only a water phase and water
diffusivity, D
w
, is known if the formation temperature and
pressure are known. Borehole temperature is a routine logging
measurement, but pressure is usually not measured directly
during logging. However, formation pressure computed from
sparsely measured and interpolated pressures, or from the
pressure gradient, is usually sufficient for D
w
estimation
because water diffusivity depends only weakly on the
pressure. The BVM fluid diffusion is unknown as it depends
on the fluid constituents. However, approximating the BVM
fluid diffusivity, , with D
BVM
D
w
will not significantly affect
the BVI estimate if any of the following apply:
a. BVM contains water only. Obviously the assumption
above is exact. WBMF invasion also belong to this
case.
b. BVM contains light oils with or without water. As
oil is the non-wetting phase, the light oil T
2
depends
little on the surface relaxation and the bulk T
2
is in
the range of several hundred ms to several seconds.
The diffusivity of light oil, D
o
, is usually in the range
of 0.1-1 times D
w
. The corresponding T
2diff,o
is on the
order of several hundred milliseconds. Thus,
replacing an unknown D
BVM
by D
w
does not
substantially shift the oil T
2
spectrum and, thus, BVM
will not be misinterpreted as BVI. OBMF invasion
also belongs to this case.
c. BVM contains medium viscosity oil, in which case
the bulk oil, T
2B
, is in the vicinity of 40100 ms at
reservoir conditions. For a typical T
2diff,w
= 100500
ms, and T
2diff,O
>> T
2diff,w
. As long as
) , min(
, diff 2 , diff 2 2 cutoff 2 w o B
T T T T << < , the discrepancy
between the water and oil diffusivity will have little
effect on the bulk oil, T
2
. Further, if one applies a
correction to the echo trains having a large GT
E
to
become one equivalent to that having a smaller or
zero GT
E
, the BVI will not be affected at all in
sandstones and only slightly in carbonates.
d. The region below the T consists, in part, of
heavy oils, in addition to the BVI. Because the heavy
oil
cutoff 2
min(
B
<< ) , and
w 2diff, o 2diff, 2 cutoff 2 2
T T T T T
B
<< , the
diffusion effect is always negligible. Thus, the BVI is
not affected.
e. Gas reservoir. Gas diffusivity is much larger than that
of liquids. With a large GT
E
, the gas signal can be
easily pushed into the BVI range. Correcting the GT
E

effect with D
w
instead of D
g
results in an insufficient
correction, but it is in the right direction to move the
gas signal out of the BVI, making the BVI more
accurate.

Therefore, we see that in the majority of situations it is
acceptable to use the known D
w
to approximate the unknown
D for the purpose of making the time-domain GT
E
correction
on echo trains. Consequently, we can combine them together
for the objective of improving the estimation of total porosity,
BVI and BVM. It is noted that the technique described in this
paper does not aim to improve the entire T
2
spectrum
estimation.

GT
E
Correction Procedure
We use the echo trains described in Figures 3 and 4 to discuss
the GT
E
correction procedure. Lets illustrate the method with
two general echo trains denoted as E
1
and E
2
. The first case
involves two echo trains having different echo train lengths. L:

( ) ( )
E E E E
T N L T N L = > =
2 1
2 1
. (7)

This represents ET #3 or 4 stacking with ET #1 in .
For such a case, the correction is applied to E
2
. Because the
highest frequency has the best SNR, in PoroPerm acquisition,
the longer echo train, E
1
, is always acquired with the highest
frequency and, thus, .
2 1
f f >
Figure 3

Step 1. Compare the summation of echoes, ( )
2
2
f S
E
and
( )
1
1
f S
E
, using the number of echoes = ; in this
case it is , using substantial vertical averaging to remove
the random noise effect. If
}
2
E
N , min{
1
E
N
.
2
E
N
( ) ( ) 0
2
2
1
1
S f S
E
f
E
consistently,
no correction is required. Otherwise, proceed to the next step.

Step 2. Apply the multiplication factor,
6 SPE 84481
( )
|
|
.
|

\
|

=
12
exp
2 2
1
2
2
2
w E
E
D T G G
kT

, (8)
to E
2
echo-by-echo, using the D
w
and G values derived from
the corresponding reservoir temperature, pressure, and
activation frequencies. The result approximates the predicted
E
2
at f
1
.

( ) ( ) (
2 1 2 2 1 2
, , , , , , , f f D T T k f E T k f E
w E E E
) . (9)

Step 3. Compare the summation of echoes, ( )
1
2
f S
E
and
, as before. If the difference is larger than tolerance,
proceed to step 4.
( )
1
1
f S
E

Step 4. If this indicates an
overcorrection. When this occurs, reduce the gradient factor,
, by a multiplication factor between 0.5 and 1 and
reiterate the process. If then an
undercorrection is indicated. When this occurs, increase the
factor by a multiplication factor between 1 and 2 and
reiterate the process. Although rarely needed, this step
provides a means to handle the over- or undercorrections that
may arise from the uncancelled portion of the internal gradient
effect described in Appendix A or from an incorrect D.
( ) ( ) ,
1 1
2 1
tolerance f S f S
E E
>
( ) (
1 1
1 2
f S f S
E E

2
2
2
2
2
1
G G
2
1
G
) , tolerance >
G
The above steps are applicable strictly to those echo trains
that are acquired with the same T
E
but different frequencies
(and, thus, gradient). This is the case of MREX PoroPerm
acquisition. The second case is more general: it involves two
echo trains that can differ in G, T
E
, and the echo train length,
L. In this case, a minor generalization of replacing S
E
with
S
E
/T
E
and replacing the multiplication factor, , by

( )
( )
|
|
|
.
|

\
|

= =
12
exp
2 2
2
2 2
1
2
2 1
1 1
w E E
E E
D T G T G
kT kT t

(10)

for the k
th
echo in an E
2
echo train at time t. For such cases, we
typically choose to correct the echo train with the largest
to be consistent to the one with the smallest
E
T G
E
T G ,
because the latter is closest to the common laboratory NMR-
measured results taken from a zero external gradient.

Time-Dependent Echo Weights
After the correction is completed, a weighted averaging
is applied to the corrected echo trains with the weights
depending on the noise and T
E
T G
E
of the individual echo trains.
The T
E
dependency is straightforward. The standard deviation
of noise for any echo in an echo train is a constant before
the G correction is applied. Because the correction is a
time-dependent factor, the noise becomes time dependent,
, after the multiplication factor, or , is applied:
E
T
E
T k t =

( ) ( ) ( ) ( ) ( ) ( ) ( ) t t N t t e t t E t M + = = , (11)

where e is the NMR signal without noise and N is the noise.
The second term shows that the noise is also amplified (or
reduced) by the same time-dependent correction factor.
To average, the corrected echo trains, a time-dependent
weight,
E
T G
( )
2
) (

= t t w , is introduced:

( )
( ) ( )
( )

=
l
l
l
l l
t
t t M
t M
2
2

, (12)

where l is the echo train index. If increases with t, the
noise of the later echoes increases, while the amplitudes of
these later echoes are small. Because the same
( ) t
l

( ) t
l
factor is
used to construct the weighting factor, the larger the ( ) t
l
, the
smaller the weight. Hence, the time dependent weight reduces
the importance of the noisy later echoes in the data integration
process.
The frequency-dependence of the noise in the original
echo trains are also taken into account with the time-
dependent weight by multiplying and then replacing with
l l l
t t * ) ( ) ( = (13)

where is the standard deviation of noise for the l
l

th
echo
train.
Further, the difference in the data sampling rate due to the
different T
E
is also built in by replacing Eq. (13) with

l
E l l l
T t t = ) ( ) ( . (14)

Equation (14) indicates that in the absence of the GT
E

correction, high-frequency and short-T
E
data are weighted
more than the low-frequency and long-T
E
data. The integrated
echo train has the S/N better than either of the constituent echo
trains.

Examples
A. Comparison of BVI and BVM from Individual and
GT
E
-corrected Echo Train Processing
Figure 6 shows a comparison of BVI and BVM estimated
from the three fully-polarized echo trains, labeled 1A, 2A, and
3A in , from the three-frequency version PoroPerm +
Oil package. The three ETs have the same acquisition
window, L, but the T
E
values are different so the N
E
s are
adjusted accordingly. Because the data were acquired with
different frequencies (See ), the corresponding Gs
differ. Thus, the additional decay ( T ) due to diffusion in
the gradient field is strongest in 1A and weakest in 2A,
causing the corresponding T
diff , 2
2
spectra to shift differently. When
a fixed T
2cutoff
(33 ms in this example) is used to compute BVI
and BVM using individual ETs, the resulting three sets of BVI
exhibits Consequently, the BVM
estimates are also affected. The trend is opposite:
Thus, to obtain consistent BVI
estimates from these three ETs, one either has to discard the
ETs with the larger , such as 1A and 3A in this case, or
integrate these ETs in the time-domain before inversion.
.
1 3 2 A A A
BVI BVI BVI < <
.
1 3 A A
BVM BVM > >
E
T G
2A
BVM
Figure 4
Table 1
SPE 84481 7
In the simplest approach, we can integrate the ETs simply
by computing a weighted average; this ignores the
difference. In this case, the weights are not time
dependent and are simply proportional to
E
T G
( )
2
2

=
l
E l l
T w .
Because 2A has the shortest T
E
, it carries the largest weight.
Therefore, the resulting integrated ET is most similar to 2A
ET. In the next example, we demonstrate that the difference in
BVI estimates between the GT
E
-corrected and uncorrected
integrated echo trains is negligible only with a very small
T
2cutoff
, but is non-negligible for a long T
2cutoff
.

B. Comparison of BVI from GT
E
corrected and non-
corrected, Integrated Echo Trains
Figure 7 shows a comparison of the BVI estimates obtained
with and without the GT
E
correction to the same ETs used in
. The same T
2
distributions were used to calculate the
BVI shown in the four tracks but the nominal T
2cutoff
value is,
from left to right, 33 ms, 40 ms, and 50 ms, respectively. As
the nominal T
2cutoff
increases, so does the discrepancy of the
results because the longer T
2
components are more affected by
the decay due to the T
2diff.

Figure 6
Figure 6. Comparison of BVI and BVM estimated from the three
echo trains, 1A, 2A, and 3A in Figure 4, in the three-frequency
version of PoroPerm+Oil package. The difference is attributed to
the GT
E
difference among these ETs.


Note that in the processing protocol described above, the GT
correction is applied to the ETs with respect to
, where is the minimal GT
combination among all acquired ETs. Ideally, the correction
should be applied with if one wants to compare
log and core data in the same condition. Usually, the
discrepancy is insignificantly small. Furthermore, if one
desires to consider the remaining small difference, one can
choose to shift the T
E
E
( ) ( )
2
min
2
E E i
GT T G
i
( )
min E
GT
( ) 0
min
=
E
GT
2cutoff
accordingly after inverting the
integrated ET. In a later example, we demonstrate that such an
additional correction is often unnecessary. The desired T
2cutoff

shift can be computed easily with the known D
w
. The T
2cutoff

shift is more desirable if the GT
E
correction is applied with
respect to an ET that has its corresponding GT
E
value greater
than ( )
min E
GT .

Figure 7. Comparison of BVI estimates obtained with and without
the GT
E
correction. The same T
2
distributions are used in the four
tracks but the T
2cutoff
value is, from left to right, 33 ms, 40 ms, and
50 ms, respectively. As the T
2cutoff
increases, so does the
discrepancy of the results because the longer T
2
components are
more affected by the decay due to the T
2diff.


C. GT
E
Correction for PoroPerm Data
The next example includes the analysis of a data set acquired
with the six-frequency version of the PoroPerm acquisition
sequence. The integrated ET includes the contribution from all
fully polarized ETs, either with a long acquisition window (L
= 500 ms) or a short one (L = 30 ms). The long ETs are
acquired with the highest frequencies which correspond to the
highest gradients; the short ETs, designed to improve the
quality of the early echoes, are acquired with lower
frequencies and, thus, the corresponding gradients are smaller.
With a typical D
w
= 810
-9
m
2
/s, the gradient difference
between the highest and lowest frequencies in the PoroPerm6
results in no more than 4.7% of the decay amplitude at t = 30
ms. Therefore, it is not necessary to consider this difference
before integrating the short and long L echo trains together.
This is particularly true if most of the fluid constituents in the
formation correspond to short T
2
components such as CBW or
BVI.
Figure 8 illustrates the six fully polarized echo trains
acquired with the PoroPerm6 sequence from Baker Atlas
Austin Test well. The leftmost plot shows all six echo trains.
The short and long echo trains have 30 ms and 500 ms
acquisition windows, respectively, but for clarity, only the first
60 ms of data are plotted from the long echo trains and all 30
ms of data for the short echo trains are plotted. The middle
panel includes the weighted average ET based on (a) only the
two long echo trains and (b) all six, long and short, echo
trains. We see that for only 30 ms of data and a T
E
of 0.6 ms,
there is no difference in the decay rate, so no GT
E
correction
is needed. The rightmost plot shows the inversion results from
the corresponding two integrated echo trains displayed in the
8 SPE 84481
center plot. A slight improvement of the spectrum
resolvability is observed by including all echo trains.



Figure 8. Illustration of PoroPerm echo train integration with and
without the GT
E
correction.

D. T
2cutoff
: Laboratory vs. PoroPerm Log
In our processing protocol, when two echo trains with
different L are integrated, the GT
E
correction is always
applied to the short L echo trains. For the PoroPerm sequence,
the short L echo trains are acquired with lower frequencies and
thus correspond to lower GT
E
values. Correcting these short L
echo trains increases the GT
E
effect. In contrast, all echo
trains in the PoroPerm sequence are acquired with a T
E
of 0.6
ms so that the maximum (GT
E
)
max
for any echo train in the
PoroPerm acquisition is smaller than the (GT
E
)
max
in
PoroPerm + Oil acquisition. To assess the significance of the
(GT
E
)
max
on the effective T ,
log
cutoff 2

( ) ( ) P T D T G T
T T E
G
, , ,
1 1 1
diff 2,
0
lab
2cutoff
log
cutoff 2
+ =
=
, (15)

and, consequently, the BVI estimates, we show a limestone
well example in Figure 9; the PoroPerm6 data were acquired
from Baker Atlass Austin test well.
In Figure 9a, from left to right, the first track shows the
total porosity, the second track shows the comparison of BVI
using the 90-ms nominal T
2cutoff
and the effective T ,
using the actual well temperature of 100 F. The BVI
difference estimated from the two T
log
cutoff 2
2cutoff
s is negligible. Even
if the well temperature were 300 F, the difference would still
be very small (Fig. 9b). The modification of the T
2cutoff

becomes desirable only if both the laboratory-derived T
2cutoff

and the well temperature are high. To illustrate this point, we
plot in Fig. 9c BVI and Perm with a modified lab-derived
T
2cutoff
of 120 ms and a well temperature of 300 F. Under
these conditions, the effect due to the T
2cutoff
shift on BVI and
permeability estimates is more noticeable. Even when it is
needed, it is a simple step to modify the T using Eq.
(15).
log
cutoff 2

E. Improvement of Repeatability
Figure 10 illustrates the improvement of BVI repeatability for
the Austin Test well using data acquired with the three-
frequency PoroPerm+Oil sequence. Three passes were
acquired in the same interval to evaluate the BVI repeatability.
The left track shows the BVI estimated from the lowest GT
E

and the fully-polarized echo train, acquired at 795 kHz. The
right track plots the BVI estimated using all long echo trains
(L=500ms) acquired at various frequencies and T
E
s. The GT
E

correction and weighting is applied to these data before
integration.
(a)

(b) (c)

Figure 9. Comparison of BVI and Permeability estimated with the
nominal T
2cutoff
and those computed from the GT
E
-corrected data.
The nominal T
2cutoff
is used here as the laboratory-derived value.
(a) T
2cutoff
(lab) = 90 ms, T = 100 F, (b) (a) T
2cutoff
(lab) = 90 ms, T =
300 F, and (c) (a) T
2cutoff
(lab) = 120 ms, T = 300 F.

The data shown in both tracks of Figure 10 have been averaged
over 2 levels. The repeatability is better with the integrated
echo trains, especially for depths where the BVI is small,
demonstrating the benefit of integrating echoes trains with
different GT
E
s.
SPE 84481 9
T
1
longitudinal relaxation time

T
2
transverse relaxation time
B
T
2
bulk fluid transverse relaxation time
T
2cutoff
dividing time between BVI and BVM
T
2diff
extra decay time factor due to diffusion
T
E
interecho time
T
W
wait time
V pore volume

Greek letters
time-dependent G correction factor
E
T
gyromagnetic ratio
the difference of
magnetic susceptibility
surface relaxivity
standard deviation of noise
time-dependent weighting factor
time-dependent G correction factor
E
T

T
total porosity

subscripts
ext external
int internal
i,j,k,l dummy indices
o oil
Figure 10. Comparison of repeatability using the best quality echo
trains (left track) and the integrated, GT
E
corrected echo train.
w water


Acknowledgments
Conclusions
Numerous individuals at Baker Atlas have contributed to the
development of the MREX tool and data acquisition and
processing software. We thank Baker Atlas for permission to
publish the results.
We have presented a method that considers the gradient and
interecho time differences between the MREX echo trains
acquired in a single-pass acquisition and thereby allows us to
integrate these echo trains in the time-domain before
inversion. The integrated echo train is used to obtain porosity,
BVI, and BVM for real-time processing without requiring real
time interpretation input from the log engineer. Furthermore,
we have discussed the T
2cutoff
discrepancy between laboratory-
based, zero-gradient NMR data and NMR log data acquired in
a gradient field. Finally, we have demonstrated that for the
MREX PoroPerm acquisition with T
E
= 0.6 ms, the
discrepancy poses negligible effect for the majority of real,
non-extreme situations.

References
1. Fang, S., Chen, S., and Georgi, D., SIMET for Fluid
Characterization: Processing Algorithms and Implementation,
BA Internal Report, 2002.
2. Slijkerman, W. F. J., Looyestijn, W. J., Hofstra, P., and Hofman,
J. P.,: Processing of Multi-Acquisition NMR Data, paper
SPE56768 presented at the 65
th
(Month 1999) ATCE of SPE,
New Orleans, Louisianna.
3. Chen, S., Beard, D. C., Gillen, M., Fang, S., and Zhang, G.:
MR Explorer Log Acquisition Methods: Petrophysical-
Objective-Oriented Approaches, Paper ? presented at
44
th
(June 2002) SPWLA Logging Symposium and
Exhibitions, Galveston, Texas.

Nomenclatures
B
0
static field strength
B
1
RF field strength

D diffusivity
Appendix: Internal Gradient Effect
e,E Echo amplitude with and without noise included
In general, fluids filling the pore space of formation rocks are
subjected to the total magnetic field gradient:
f frequency
G RF field gradient strength
int
G G G
ext total
r r r
+ = ,
(a1)
which is the vector sum of the external and internal gradients.
The internal gradient arises from the differences in magnetic
susceptibility between the matrix and the saturating fluid.
Generally, it depends on both pore geometry and the type of
rock chemical constituents (mineralogy). The internal gradient
can, in principle, be significant. It is proportional to the
external field strength:
L echo length, N
E E
T
M Echo magnetization amplitude
N
E
number of echoes in an echo train
pore
r pore radius
S pore surface
S
E
Sum of echoes
t time
T temperature
10 SPE 84481
The effect of the internal gradient in Eq. (a4) can be embedded
into the rest of terms:
pore
0
int
~
r
B
G

, (a2)
1
ext 2diff,
1
int 2diff,
1
2
1
2

+
(

+ + = T T
V
S
T T
B
. (a5)
thus, it is frequency dependent and represents the
difference in the magnetic susceptibility between the fluid and
the solid matrix, and r
pore
represents the effective radius of the
pores). Variation in the operating frequency of the current
NMR logging tools, the variation of B
0
is less than a factor of
2 and so does the internal field gradient. The variation of the
external gradient is greater. More importantly, the time and
ensemble averaging of the randomly orientated G
int
eliminates
the term
ext int
G G . The phase displacement is proportional
to the sum of the squares of the internal and external gradients
The term in the brackets in Eq. (5a) is independent of the
external gradient. The internal gradient strength variation is
linearly proportional to the frequency variation as
f B G
0 int
,
while the external gradient variation is greater than the linear
proportionality of f. For most cases where on the average
int
G G
ext
>>
and T does not dominate the term in the square bracket,
we approximate that the quantity in the square bracket of Eq.
(a5) is frequency independent. Thus, if all the data to be
combined have the same T
1
int 2diff,

E
, the term in brackets is regarded to
be approximately the same for all echo trains. If T
E
values of
these echo trains are different, the term in the bracket is
approximately valid only if the decay term due to internal
gradient is much smaller than that of the bulk and surface
relaxation terms combined or is much smaller than the
external gradient term. These conditions are practically and
approximately valid in many cases.

( )
2
ext
2
int ext int
2
ext
2
int
2
2 G G G G G G G
total
+ + + =
r
. (a3)

Thus, the diffusion decay rate, Eq. (a3), which is a function of
the gradient, can be separated into the internal and external
gradient, respectively:

12 12
2 2
ext
2 2 2
int
2
1
ext 2diff,
1
int 2diff,
1
diff 2,
fluid E fluid E
D T G D T G
T T T

+ = + =

.

(a4)