Note: Descriptions are shown in the official language in which they were submitted.
449~
F-1946 -1-
A SEISMIC EXPLORATION SYSTEM AND AN
ANALOG-TO-DIGITAL CONVERTER FOR USE THEREIN
This invention relates to a seismic exploration system and an
analog-to-digital converter or use therein.
- For many years it has been common to explore for oil, gas and
other valuable minerals using seismic techniques. seismic energy is
imparted to the earth by, for example, detonating a "shot" of dynamite
in a hole on the earth's surface or by imparting a mechanical vibration
to the earth. The wave reflects from interfaces in the earth's crust,
and is detected by detectors spaced some distance from the point at
which the seismic energy is imparted to the earth. The signals output
by the detectors are recorded. By measuring the time taken by the waves
to travel over plural paths to plural detectors, conclusions can be
reached about the shape of interfaces separating varying rock layers in
the earth's subsurface formation. From analyses of these interfaces,
likely locations for deposits of oil, gas and other valuable minerals
can be identified.
A perennial problem in the accurate measurement of the time
taken by the waves is recording the signals with a sufficiently good
signal-to~noise ratio to enable the received waves to be reliably
distinguished from noise occurring in the earth and generated by the
exploration process itself. In particular, when marine seismic
exploration is performed, acoustic microphones, referred to hereinafter
as "hydrophores", are trailed behind a seismic exploration vessel. The
vessel includes means for imparting an acoustic wave to the ocean, which
then travels through the ocean and into the sea bed. The wave is
reflected from the interfaces between the rock layers forming the sea
bed, back up to the hydrophores streamed behind the exploration vessel.
The return of the reflected wave is detected by the hydrophores, which
typically output an analog voltage signal. Typically, the signals from
the hydrophores are sampled and a digital representation of the
instantaneous amplitude is stored for later analysis.
~204~4
F-1946 -2-
The analysis of seismic signals is growing increasingly
sophisticated and it is clearly desirable to provide as accurately
recorded signals as possible. To this end, there has for some time been
a need in the art to eliminate extraneous signals, or "noise", from the
recorded digital samples. The term "noise" is used in geophysical
research to indicate any unwanted signal; this can be clarified by
noting that such unwanted signals can be generated by numerous sources.
For example, in marine seismic exploration acoustic waves generated my
other vessels in the vicinity of exploration, or by turbulence generated
by the streamer itself, contribute "noise". Imprecise electronic
components and inaccurate digitization processes add "noise". Signal
degradation also occurs during transmission of the signals from the
hydrophores up the "streamer" cable to the exploration vessel for
recording, effectively reducing the signal-to-noise ratio. It would
obviously be desirable to improve the signal-to-noise ratio of such
marine seismic explorations by any means possible so as to allow better
identification of geologically significant events in the seismic record.
Those skilled in the art will recognize that not all of these
classes of noise can be removed from the seismic record using a single
technique. For example, the coherent noise generated by other vessels
in the exploration region or by turbulence can be eliminated by
mathematical filtering, but this has less effect on incoherent noise,
such as that added to the seismic record by inaccurate digitization.
The present invention addresses the problem of "noise" caused
by inaccurate digitization processes.
One prior art analog-to-digital conversion technique which has
been employed in seismic applications uses analog devices known as
galn-ranging amplifiers in the signal path. Such amplifiers are capable
of amplifying both small and large signals by varying ratios so that
quite small signals can be effectively digitized; that is, such
gain-ranging amplifiers have good dynamic range characteristics.
However, the gain of the amplifier with respect to any particular input
signal amplitude is fixed and hence if a very small signal is
superimposed on a very large signal, the small signal will not be
4~94
F-1946 -3-
amplified adequately to be digitized and will be lost. Accordingly,
such gain-ranging amplifiers, lacking adequate "resolution", introduce
substantial distortion into certain classes of input signals by randomly
truncating small signal values. Furthermore, gain-ranging amplifiers
are nonlinear analog devices which introduce distortion of the absolute
amplitude of the input signal which is highly undesirable.
It has been determined that no presently available
analog-to-digital converter, including the gain-ranging amplifier types
discussed above, is available in the prior art having an adequate
signal-to-noise ratio, while providing adequate resolution of signals of
widely varying amplitudes, to fully utilize present day data processing
capabilities. It would be desirable to accurately record signals of up
to approximately 120 dub level difference for seismic analysis. The
present invention is designed to provide such dynamic range in encoding
analog signals, while providing accurate resolution of a small signal
superimposed on larger signal values. In particular, it minimizes the
"quantization noise" introduced by all analog-to-digital conversions,
and avoids distortion caused by use of a nonlinear analog device such as
a gain-ranging amplifier.
Accordingly, the invention resides in one aspect in a seismic
exploration system comprising:
a source of seismic energy;
a detector of reflected seismic energy comprising means for
outputting a detected analog signal;
means for converting the detected analog signal to a digital
signal; and
means for recording the digital signal;
wherein the means for conversion of the detected analog signal
to a digital signal comprises:
means for predicting an expected input signal and for
generating a digital representation thereof;
digital-to-analog conversion means for conversion of the
digital output of the predicting means to a predicted analog signal;
12~4~94
F~1946 I
means for comparing the output of the digital-to-analog
conversion means to the detected analog signal;
means for conversion of the output of the means for
comparing to a digital difference signal representing the difference
between the detected analog signal and the predicted analog signal;
means for adding the digital difference signal to the
digital predicted value; and
means for transmission of the sum of the digital
difference signal and the digital predicted value to the recording means.
In a further aspect, the invention resides in an
analog-to-digital converter for use in a seismic exploration system
comprising:
means for generating a predicted digital input signal;
means for conversion of the predicted digital input signal to
an analog representation thereof;
means for comparing the predicted analog input signal with an
actual analog input signal, and for generating an analog error signal
representing the difference there between;
means for converting the analog error signal to a digital error
signal representative thereof;
means for summing the digital error signal and the digital
predicted signal;
means for transmission of the sum of the digital error signal
and the digital predicted signal to means for recording the summed
digital signal; and
means for updating the predicted analog input signal based on
the sum of the digital error signal and the digital predicted signal.
In a preferred embodiment of the invention, over sampling is
applied to the relatively low frequency seismic signals so as to provide
many more samples than would strictly speaking be necessary for
representation of the simple seismic waveform. Summing over the many
samples provides a smoothing effect to the data and thus provides
further accuracy. It also enables the predicted values to be
~04~
F-1946 I
synthesized with greater accuracy, such that the error signals are
smaller, and may be represented using smaller digital words for a given
level of accuracy. Alternatively, better encoding accuracy is achieved
with a word of a given length used to digitize the smaller error
signals. Either approach results in reduced quantization noise.
In the accompanying drawings which illustrate one example of
the invention,
Figure 1 shows a marine seismic exploration system in a
schematic form;
Figure 2 is a block diagram form of the analog-to-digital
converter of the system; and
Figure 3 is a block diagram of a hardware implementation of the
circuit shown in Figure 2.
Referring to the drawings, Figure 1 shows a marine seismic
exploration system in which an exploration vessel 10 tows behind it a
streamer cable 12 comprising a plurality of hydrophores 14. A source of
seismic energy 16 on the vessel, which may be a compressed air gun or
the like, transmits seismic energy down various ray paths 18 to be
reflected at the ocean bottom 20 or from an interface 22 between varying
rock layers of the subset bed and reflected back upwardly along
differing ray paths 24 to be received by hydrophores 14. The analog
signals received by the hydrophores 14 are converted into digital
signals by analog-to-digital conversion means 26 prior to being recorded
on recording device 28.
The present system uses the analog-to-digital conversion
technique known as differential pulse code modulation to convert thy
analog signals output by the hydrophores 14 into digital signals.
According to this technique a prediction operator is used to compute a
predicted signal which effectively models the anticipated seismic analog
signals to be received. The actual analog signal received is then
compared with the predicted signals at regular intervals; the difference
between the two or the "error" signal is digitized and added to the
digital predicted value for transmission.
~21~4~
F-1946 -6-
The predicted value is continually updated using the past
predicted values and the error value. This coding method offers
substantial advantages with respect to the signal-to-noise ratio ox the
conversion and the resolution which can be achieved in the process.
Thus, the dynamic range of the signals which can effectively be recorded
is broadened. Differential pulse code modulation is particularly
effective for seismic exploration, where the signal to be encoded is a
relatively accurately predictable signal such as a low frequency
sinusoid. This allows the predicted signal to match up fairly
accurately with the received signal so that the error is relatively
small. As noted, a relatively small error value can be accurately
digitized in a relatively small number of digital bits, thus effectively
reducing the quantization noise introduced in any analog-to-digital
conversion. By comparison, a large error signal is less accurately
digitized in a digital word of a given size; to accurately digitize a
large value, e.g. a larger error signal or the actual analog value, a
longer digital word must be generated all with attendant complexity and
expense.
As shown in Figure 2, the differential pulse code modulation
system includes a dif~erencing junction 30 at which the continuous
analog input signal arriving from the hydrophores 14 along input line 42
is compared at intervals with a predicted analog signal generated by a
digital-to-analog converter 44 based upon a digital predicted value
output by linear predictor 40. Accordingly, the output of the
differencing junction 30 is the analog difference between the actual
analog signal and the predicted analog signal, that is, the analog error
signal. The analog error signal is then passed to an amplifier 32 for
scaling and converted back to a digital signal in an analog-to- digital
converter 34. It will be appreciated that this analog- to-digital
converter 34 operates on the error signal which has a much smaller
dynamic range than the analog input signal. Accordingly, the
analog-to-digital converter 34 need not be capable of the wide dynamic
range of the entire system shown in Figure 2. For a given level of
accuracy of digitization, the digital words output by converter 34 may
.
4~914
F-1946 -7-
thus be shorter, i.e. comprise fewer bits, than the digital words output
by the digital predictor 40, which correspond to the actual input
signal. Accordingly, the signal output by the converter 34 is
reformatted as necessary in a formatter 36, in which the value of its
least significant bit is scaled to match that of the longer word output
by predictor 40, before being added thereto in junction 38 and
transmitted. Successive sums of the predicted value and digitized error
signal are then digitally filtered in a low pass filter 46 and
sub sampled by a sampling unit 48 before becoming the output signal,
which would comprise digital "words" representing the instantaneous
value at the sampling time. The operation of the low pass filter and
sub sampler Jill be discussed in further detain below. ye sum of the
predicted value and the error signal also goes to the linear predictor
40 for updating ox the predicted value in accordance with the error
signal, as noted. In this way the predicted signal is continually
varied in accordance with the actual signal so as to constantly provide
an updated and hence increasingly accurate prediction. This cycle is
then repeated: the output of the prediction operator is transformed into
an analog signal in digital-to-analog converter 44, is passed to the
differencing junction 30 where the analog signal is again compared
thereto; the analog difference is digitized, added to the predicted
value for transmission and for use by the linear predictor to generate
the next predicted value.
As discussed above, it is desirable that the error signal be as
small as possible, so that for a digital horn of given length output by
the analog-to-digital converter 34, the error signal can be most
accurately represented. In order that this be the case, it is Obviously
necessary that the predicted value output by the prediction operator
equal the input signal as accurately as possible. This, in turn,
requires that the sampling rate of the input signal used to update the
linear prediction operator be chosen so as to ensure that an adequate
number of samples are generated. In theory, any analog signal can be
adequately encoded if it is accurately sampled at twice the highest
frequency signal making up the analog signal, i.e., at the Nyquist
~204~94~
F-1946 -8-
rate. In seismic exploration, the input analog signals are typically
band limited to approximately 125 Ho. Accordingly, Nyquist's theorem
would indicate that a 250 Ho sampling rate would be adequate. It will
be appreciated that this is far below the abilities of modern
electronics. Preferably, the sampling rate is raised to much higher
than the Nyquist rate. This "over sampling" which may occur at e.g., 6û
kHz, increases correlation between samples which enables better
prediction. The remaining noise can be reduced with respect to the
signal by low pass filtering (smoothing). Quantization noise is thus
reduced, as the least significant bit of the digitized error signal is
smaller. Low pass filter 46 smooths the output signal; this in
conjunction with sub sampler 48 which ensures that the output signals -
which would typically comprise digital words, output at or near the
Nyquist frequency - accurately represent the received analog signals.
It will be appreciated by those skilled in the art that the
input signal can only be adequately represented by the predicted signal
if the bandwidth of the loop is wide enough to encompass the fastest
signal swings undergone by the input signal. In order that this can be
so in all cases, it may be desirable in some cases to incorporate an
integrator 34 and a clipper 50 in the input path. The output of
integrator 34 will change with respect to the rate of change of the
amplitude of the analog input signal. If a clipper 50 is used, the
slope ox the output of integrator 34 can be limited to the loop's
bandwidth. To ensure that the output signals still represent the
integrated analog input signals accurately, a differentiator 51 should
be interposed in the output signal path.
Figure 3 shows a schematic diagram of a circuit which could
perform the functions shown in Figure 2. Input signal K is input to
a low noise amplifier 52 which performs the function of comparison of
the input signal with a predicted signal v output by an 18-bit
digital-to-analog converter 54. The analog output error signal f is
input to a 12-bit analog-to-digital converter 56 which outputs a Betty
digital error signal. This is reformatted in a Betty output ROM 58 so
as to scale the least significant bit of the 12-bit error signal to
~0449~
F-1946 I
equal that of the 32-bit predicted input signal. As well understood by
those skilled in the art, the 12-bit error signal serves as the address
to the ROM 58, and the data stored in the ROM 58 corresponds to the
32-bit version of the error signal. This 32-bit error signal is one
input to an adder 60, the other input of which is an 18-bit truncated
version (the 14 least significant bits being lost) of the 32-bit
predicted signal. The output of the adder 60 is the sum of the 32-bit
error signal and the truncated 18-bit version of the 32-bit output
signal. This signal is then input to a first shift register 62 and
becomes one of three 32-bit inputs to a sign multiplier with accumulator
64 which performs the function of the linear predictor 4û Chilean in
Figure 2. In successive clock cycles of the circuit shown in Figure 3,
the 32-bit value in shift register 62 is successively passed to shift
registers 66 and 68, so that each 32-bit value is input three times to
the sign multiplier and accumulator 64. This is also supplied with
weighting values corresponding to the linear predictor designed into the
circuit by three ROMs 7û9 72, 74. The 32-bit output signal, which is
the predicted value of the analog input signal, is passed to a formatter
76 where it is truncated to provide the 18-bit input to the labia
digital-to-analog converter 54 and to the 32-bit adder 60. Truncation
is performed because current single chip digital-to-analog converters
are not available which operate on 32-bit input signals.
The output of the 32-bit adder 6û forms the output ox the
circuit, which is operated upon by digital filter and sub samplers 46 and
48 shown in Figure 2 to provide digital words forming the output of the
circuit.
The key hardware items of the diagram shown in Figure 3 are the
digital-to-analog converter 54 and the analog-to~digital converter 56.
Adequate devices are presently available in the market. An embodiment
of the invention corresponding to Figure 3 was simulator tested by
computer operation. The specifications ox an 18-bit digital-to-analog
converter made by Analog devices and sold under model number 1138K were
used in the simulation for digital-to-analog converter 54. Similarly,
the hypothetical analog-to-digital converter 56 used in the simulation
120~494
F-1946 -10-
was Analog Devices' Model 110~-00~ converter, which outputs a 12-bit
digital representation of the error signal.
The linear predictor 40 used in the simulation was a 1 to 3
coefficient linear predictor, envisioned to be implemented with 32-bit
fixed point hardware. The coefficients can be based on a minimum mean
square prediction error criterion using a band limited flat input signal
model. The design of such linear predictors is fully discussed in
Rabiner & Schafer, "Digital Processing of Speech Signals," Prentice
Hall, (1978). In the simulated embodiment, the coder operated at a
frequency of 60 kHz. This extremely high sampling rate compared to the
signal bandwidth of O to 1~0 Ho resulted in an adjacent sample
auto-correlation of 0.99998~. An alternative would be to use a single
coefficient predictor with dithering to ensure non correlation of the
error values.
The predictor output was a 32-bit prediction in the embodiment
simulated, which was truncated as noted to 18 bits for input to the
digital-to-analog converter 54. Similarly, the 12-bit error signal must
be reformatted to 32 bits before being input to the linear predictor.
As suggested, such reformatting could be readily accomplished using
ROMs, with the 12-bit error signal serving as the address, and the
32-bit input to the predictor being the data. Alternatively, a digital
multiplier could perform the reformatting function; the multiplication
ratio of the reformatting operation would depend on the ratio of the
values of the least significant bits of the output of the 32-bit
predictor 64 and the 12-bit analog-to-digital converter 56.
The low pass filter I in the simulated embodiment was a
two-stage finite impulse response non-recursive digital filter as
described in Rabiner and Gold, "Theory and Application of Digital Signal
Processing", Prentice Hall, (1975). Filter parameters tested were as
follows. First stage N = filter order = 1001, Us = sampling rate - 60
kHz, Fc = break frequency = 1200 Ho. Second stage, N = 501, Fc =
~000 Ho, Fc = 150 Ho.
The simulation results indicated that the total signal to
root-mean-square quantization noise ratio equals or exceeds 126 dub for
seismic band signals.
