GB2339094A - Analog to digital converter - Google Patents

Description

2339094 A SPECTRAL ANALOG-TO-DIGITAL CONVERTER This application is a

continuation-in-part of Application No., filed July 1, 1998, entitled "A POST-CONVERSION SYSTEM FOR IMPROVING ANALOG-TO DIGITAL CONVERTER RESOLUTION" [010262-0096001.

BACKGROUND OF THE INVENTION

Analog-to-digital converters (ADC) convert an analog voltage into a binary representation in the form of digital,.,:

bits. The resolution of the ADC is expressed in the number of digital bits into which the analog voltage is converted.

Generally, the higher resolution that is desirable, the more precision required in the components of the analog-to-digital converter. For example, amplifying transistors, etc. need to have tighter tolerances and be very precise for enabling a meaningful least significant bit resolution.

In imaging applications, it is desirable to improve the image quality. one method for doing this is to use a boost filter at the output of a subsystem. However, since boost filters also amplify noise, they are typically used in high contrast situations, and not applications with a low signal-to-noise ratio.

SUMMARY OF THE INVENTION

The present invention recognizes that for visual systems, a receptor (such as the human eye) is most sensitive in only certain spacial frequency ranges. A frequency range of improvement in the ADC signal-to-noise ratio is matched to the frequency range of highest sensitivity in the receptor.

Accordingly, an analog to digital conversion of high resolution and accuracy is provided only for such a frequency range to which the eye is most sensitive. Frequencies outside this range are converted into digital form with a lesser resolution and accuracy. This results in the signal-to-noise 2 ratio (SNR) being improved for the spectrum of interest.

Thus, a high accuracy conversion is only needed for a portion of the spectrum to effectively increase the perception of image quality.

The present invention, in one embodiment, improves the signal-to-noise ratio on the input side of a system, rather than the output side. Analog signals corresponding to a sensed image are generated. The spectrum will vary for different areas of the image. The spatial frequency will vary for those analog signals compared to other analog signals.

All the analog signals are converted into digital form for processing, but those analog signals in the spatial frequency range of interest are converted with higher resolution and accuracy. This lowers the noise floor for the range of interest. The frequency may be based on the relationship to other samples in a horizontal line, a vertical line, or for an area.

In one embodiment, the frequency range of the high accuracy conversion may vary for different colors.

Alternately, some colors may have high accuracy conversion while others are totally lower accuracy conversion.

In one embodiment, an analog filter is used to select the data which will receive the high accuracy conversion. In another embodiment, post processing of the converted digital bits from an ADC is used to provide one or more additional bits of resolution.. For example, a digital filter is used to interpolate the extra bit from adjacent samples from the output of the ADC.

For a further understanding of the nature and advantages of the invention, reference should be made to the following description taking in conjunction with the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1A is a diagram illustrating the prior art spectral response of an ADC. Fig. 1B is the spectral response according to the present invention.

3 Fig. 2A is a graph illustrating a logarithmic plot of modulation threshold vs. spatial frequency; Fig. 2B is a graph illustrating modulation transfer function vs. spatial f requenc-,,-. -T Fig. 3 is a block diagram of an example embodiment of the invention.

Fig. 4 is a graph illustrating the linear interpolation and threshold of one embodiment of the invention.

Fig. S is a flow chart illustrating one embodiment of the method of the invention.

Fig. 6 is a block diagram of one embodiment of a circuit for providing two additional bits of resolution according to the invention.

Fig. 7 is a block diagram of one mode of the invention for a single color or monochrome luminance.

Fig. 8 is a block diagram of the second mode of the invention for multiple colors or different spatial frequencies.

DESCRIPTION OF THE PREFERRED EMBODIMENT

Fig. 1A illustrates the analog-to-digital converter (ADC) signal to noise ratio (SNR) vs. frequency (f) as a curve 100. As can be seen, the SNR is approximately 66 until very high frequencies, being essentially constant across the range.

Fig. 1B illustrates the graph of Fig. 1A applied to the present invention. As can be seen, the SNR starts out at a lower value of about 60 in a portion 102, then peaks to a value of 66 in a portion of 104, returning to a lower value of 60 in a portion 106. The higher SNR portion 104 corresponds to the frequency range that is most sensitive to the human eye. Thus, the physical implementation can provide the additional accuracy only where it is most needed.

Fig- 2A illustrates the modulation threshold (Mt) 3S based on the human vision svstem (HVS) model. This illustrates the eye's minimum detection capability of a perceived sinusoidal target. The Mt is typically J-shaped and most sensitive to spatial frequencies between 3 to 5 cycles 4 (cy) /degree (deg) at typical ambient light levels. 'Mt is defined as the ratio of the difference of the maximum and minimum luminance to the sum of the maximum and minimum luminance, The contrast sensitivity of the eye is the inverse of the modulation threshold curve. The normalized contrast sensitivity is the modulation transfer function (MTF) of the eye shown in Fig. 2B. This curve shows the range of spatial image frequencies to which the human eye will be most sensitive to image sharpness.

The present invention matches the frequency range of Fig. 23 to the SNR of Fig. 1B. This results in an optimum ADC for human vision, while minimizing the resolution and accuracy needed for frequencies to which the eye is less sensitive.

Fig. 3 is a block diagram of one embodiment of te" is invention. A light source 160 illuminates a target scene 1G2. Reflected light is directed by a lens system 164 to an imager 166. The imager provides a serial waveform in raster scan format for a horizontal line. The signal is processed by an analog conditioning circuit 112, which would typically include a correlated double sampler (CDS). The output is an analog waveform having the frequency corresponding to a current line of the image. This output is then provided to a spectral ADC 114, consisting of an ADC 116 and a time average interpolation circuit 118. Spectral ADC 114 has an SNR vs. spatial frequency as set forth in Fig. 1B.

The output of spectral ADC 114 is provided to digital post process circuitry 168, and then to an image display 170. The image is viewed by a receptor 172, which may be the human eye. Alternately, a receptor could be a photosensor used in a robotic system, for example. The MTF vs. spatial frequency response of the receptor, such as shown in Fig. 2B, is matched to the SNR vs. spatial frequency response of the spectral ADC 114.

In one embodiment, the present invention only addresses the spatial frequency in the horizontal direction to be cost effective. Typical image sensors shift image data out horizontally to be processed. Most video and scanning system sensors (CCDs, contact image sensors, MOS imagers) shift serial image data on a horizontal axis, for example. However, the invention provides sufficient improvement since the human eye is more sensitive to horizontal aberrations. A break or discontinuity which runs horizontally (indicating an error in the vertical direction) is more acceptable to most humans, because they accept the aliasing in commercial televisions -which have periodic horizontal lines distorted by raster scanning.

The same technique can be applied in the vertical direction to increase resolution. Subsequent horizontal lines are stored in the digital domain and additional resolution is generated. This is particularly applicable to arrays that can be randomly addressed as opposed to serially addressed. An example of horizontal serial shifting is the traditional CCD area array imager that shifts data our serially one horizontal line at a time from a single output. An example of a randomly addressable array is a MOS imager with individually addressable pixels that could be shifted out in a number of ways.

Example Interpolation Circuit Fig. 4 illustrates by arrows 10, 12 and 14 three different sequential outputs of an ADC. The order of the samples is (n-1), n, (n+1). The value of each of these samples are, respectively, S(n-1), S(n), and S(n+l). In one embodiment of the invention, it is -assumed that the actual analog value is varying linearly, and accordingly a line 16 is drawn between the values of the (n-1) and the (n+l) samples.

The point of intersection where the n sample appears is labeled A.

A threshold 18 is set, and the value S(n) is compared to see if it is within the threshold. Preferably, an extra bit of resolution is set to 1 only if S(n) is within the threshold.

If S(n) is greater than A, adding another bit would only make it farther from expected straight line, and accordingly the extra bit would be set to 0. If S(n) is less than threshold 18, and threshold 18 is 1/2 LSB, it is clear 6 that the actual sample is varying from the expected correlated value, and accordingly the additional bit of resolution is set to 0.

-hod for Fig. 5 is a flow chart illustrating one met.

S doing the calculation as described in Fig. 4. First, the value of A is determined as the average of S(n-1) and S(n+l) in step A. Next (step B) the actual value of S (n) is compared to the lower end of threshold 18 of Fig. 4, to determine if it is greater than.1/2 LSB below A. If it is not, the additional bit, the LSB, is set to 0 (step C).

If S(n) is greater than this lower end of the threshold 18, is compared to see if it is less than A (step D). If it is not, then it is greater than the upper end of threshold 18, the LSB is set to 0 in step C. Otherwise, th'd LSE is set to 1 (step E) indicating that S(n) is within threshold 18.

Fig. 6 is a block diagram of one embodiment of logic for implementing the present invention. A 10 bit ADC 20 receives an analog input on line 22 and provides a 10 bit output [S(n-1)) on line 24. Previous outputs are stored in registers 26 and 28. Register 28 stores the S(n+l) value, while register 26 stores the S(n) value.

A summing circuit 30 sums together the S(n+l) and S(n-1) values. Because two 10 bit numbers are summed, the result can be an 11 bit number with carry. The upper 10 bits, including any carry, are provided to a comparator 32 on lines 34. These are compared to the S(n) bits on line(s) 36. By doing the sum and taking off Zhe 10 most significant bits and shifting them down one, the result on line 34 is effectively the sum, divided by 2, of S (n+1) and S (n- 1) If the values are not determined to be equal in comparator 32, this indicates that S(n) is not within the threshold of one-half LSB, and a control signal on line 38 causes a zero value from switch 40 to be provided as the additional bit of resolution on line 42.

If S(n) is determined to be equal to the result on line 34, the value of S(n) may be within the threshold and thus line 38 is activated to selected the LSB on line 44. if 7 the LSB is 1, this indicates that the average divided by 2 is slightly high than S (n), and thus the one bit is passed through to line 42 to move the value S(n) closer to the average. If the value LSB on line 44 is zero, this indicates that the average is not greater than S (n) to this level of resolution, and accordingly, S(n) should not be incremented. Thus, the zero on line 44 is passed through to line 42 as the additional bit. The result is 11 bits of resolution on lines 46.

To produce another bit of resolution, the entire process can be repeated using a second set of registers 48, 50, and another summing circuit 52. In addition, another conparator 54 is used along with another switching circuit 56. In some applications, it may be desirable to add 'a' noise component to the additional bit of resolution when that bit is not within the threshold. This might be beneficial for a noisy environment, for instance. It may help with some calibration programs, making it easier for them to converge. Accordingly, Fig. 6 shows an optional dither circuit 58 controlled by optional control line 60 from comparator 54. When control line 60 indicates that the zero should be selected, instead the circuit 58 is'selected to place a pseudo random number (1 or 0) on line 62 to give the twelfth bit in the output on lines 64. A dither circuit 59 could also be used for the lith bit of resolution, if desired.

This embodiment of the present invention thus provides an improved resolution ADC which does not require the high precision components of a standard ADC of the same resolution. It has been determined that the maximum signal degradation is one code out of 2n. The invention can be realized using less silicon area than an equivalent implementation using analog components to directly generate the extra bits of resolutions from the analog signal. The invention improves the signal to noise ratio of ADCs, for low frequency signals, without significantly degrading highfrequency signals. As noted above, these benefits are obtained for locally correlated signals, such as may be encountered in imaging data signals.

8 Fig. 7 is a diagram of one implementation of a single color mode of the present invention. Analog conditioning circuitry 120 provides a signal to an ADC 122. A signal is then provided to a storage circuit 124 where it can be accessed by interpolation circuitry 126.

Fig. 8 shows an alternate mode for three different colors. The three colors can be input on separate analog conditioning circuits 128, 130, and 132. The signals are then provided through programmable gain amplifiers (PGA) 134, 136, and 138 to a multiplexer 140. The color signals are alternately provided through the same ADC 142, and then separated to separate storage circuits 144, 146, and 148, where they will be accessed for interpolation by interpolation circuits 150, 1 52, and 154.

Alternately, other implementations are possible.

For example, a Bayer pattern can be provided to sample CDS circuits or a sample and hold circuit, and then provided through a PGA and MUXed into a single ADC. Alternately, G and B/R colors can be simultaneously sampled, and provided to a sample and hold or CDS, and then subsequently provided through a PGA to a multiplexer into a single ADC. Another example could provide a triple color serial output sample and hold or CDS, which is provided through a PGA and multiplexed into a single ADC. Other implementations will be obvious to those of skill in the art.

The present invention can also be provided to give increased vertical resolution and accuracy by interpolating additional accuracy from vertically adjacent image data.

However, this embodiment requires more circuitry to store the image data.

In one embodiment, the range of frequencies for which higher resolution is provided could be varied for each color. For example, the eye is typically less sensitive to variations in blue, and thus a narrower range of blue freauencies could be provided higher resolution. Alternately, all of the blue color could be provided at a low resolution, while green and red, for example, are complet,ely or mostly provided at a higher resolution.

9 Alternate embodiments for providing the spectral response of the invention could be used. For example, a bandpass filter could be placed in front of a high resolution ADC. Alternately, a band-pass digital filter could be used after an ADC. For example, a 10-bit wide signal path could be used for the low resolution, switching to a 12-bit wide path for high resolution.

As will be understood by those skilled in the art, the present invention could be embodiment in other specific forms without departing from the essential characteristics thereof. For example, when an extra bit of resolution is determined for an S(n) signal, that extra bit could be used when that same signal is the subsequent S(n+ l) signal, and is being used to determine the average. This could be used is Ln alternate method to generate two additional bits of resolution at one time, rather than repeating the steps for obtaining the first bit of resolution. In other embodiments, curves other than a straight line could be used, and using more than two samples to approximate such a Lurve could be involved in the calculation. Alternately, other threshold values could be used. Accordingly, the foregoing description is intended to be illustrative, but not limiting, of the scope of the invention which is set forth in the following claims.

Claims (23)

1. WHAT IS CLAIMED IS:

   1 1. A method for performing analog to digital 2 conversion, comprising:

   3 receiving a plurality of samples, each sample having 4 a frequency relationship with adjacent samples; determining a first group of samples having a first 6 relationship with respect to adjacent samples; 7 determining a second group of samples having a a second relationship with respect to adjacent samples; 9 converting said first group of samples into digital 1 Q form with a first level of resolution and accuracy; and 11 converting said second group of samples into digital 12 form with a second level of resolution and accuracy, said 13 second level of resolution and accuracy being less than 14 said first level of resolution and accuracy.

   1
2. 2. The method of claim 1 wherein:

   2 said first group of samples is in a first frequency 3 range with respect to adjacent samples; 4 said second group of samples is in a second frequency range with respect to adjacent samples, said 6 second frequency range being less discernable to a 7 receptor than said first frequency range.

   1
3. 3. The method of claim 1 further comprising:

   2 collecting statistical information for an image 3 identifying a portion of the image having said first 4 relationship, said first relationship being where the most information is contained in said image; and 6 said identified portion being said first group of 7 samples and the remaining portion of said image being 8 said second group of samples.

   1
4. 4. The method of claim 1 wherein said recentor is a 2 human eye.

   1
5. 5. The method of claim 1 wherein said first 2 frequency range is varied for different colors.

   1
6. 6. The method of claim 1 wberein said f irst group 2 of samples comprises a first color, and said second group of 3 samples comprises a second color.

   4
7. 7. The method of claim 1 wherein said samples are 6 interleaved, and said converting steps are performed on a time 7 multiplexed basis.

   1
8. 8. The method of claim 1 wherein said first and 2 second relationships are determined for only a first dimension 3 of an image.

   4
9. 9. The method of claim 8 wherein said first and second relationships are also determined for a second 6 dimension of said image.

   1
10. 10. The method of claim 1 wherein said step of 2 converting said first group of samples into digital form with 3 a first accuracy comprises; 4 (a) determining a curve based on at least two digital samples; 6 (b) determining whether an intermediate sample is 7 within a threshold of said curve; and 8 (c) setting an additional bit of resolution based on 9 whether said intermediate sample is within said threshold.

    1
11. 11. An apparatus for performing analog to digital 2 conversion, comprising; 3 an input configured to receive a plurality of 4 samples, each sample having a frequency relationship with an adjacent sample; 6 means for determining a first group of samples 7 having a first relationship with respect to adjacent 8 samDles; 12 9 means for determining a second group of samples having a second relationship with respect to adjacent 11 samples; 12 a converter configured to convert said first group 13 of samples into digital form with a first level of 14 resolution and accuracy and to convert said second group of samples into digital form with a second level of 16 resolution and accuracy, said second level of resolution 17 and accuracy being less than said first level of 18 resolution and accuracy.

    1
12. 12. The apparatus of claim 11 wherein:

    2 said first group of samples is in a first frequency 3 range with respect to adjacent samples; 4 said second group of samples is in a second frequency range with respect to adjacent samples, said 6 second frequency range being less discernable to a 7 receptor than said first frequency range.

    1
13. 13. The apparatus of claim 11 further comprising:

    2 means for collecting statistical information for an 3 image identifying a portion of the image having said 4 first relationship, said first relationship being where the most information is contained in said image; and 6 said identified portion being said first group of 7 samples and the remaining portion of said image being 8 said second group of samples.

    1
14. 14. The apparatus of claim 11 wherein said receptor 2 is a human eye.

    1
15. 15. The apparatus of claim 11 wherein said first 2 frequency range is varied for different colors.

    1
16. 16. The apparatus of claim 11 wherein said first 2 group of samples comprises a first color, and said second 3 group of samples comprises a second color.

    13 1
17. 17. The apparatus of claim 11 wherein said samples 2 are interleaved, and said converting is performed on a time 3 multiplexed basis.

    1
18. 18. The apparatus of claim 11 wherein said first 2 and second relationships are determined for only a first 3 dimension of an image.

    1
19. 19. The apparatus of claim 18 wherein said first 2 and second relationships are also determined for a second 3 dimension of said image.

    1
20. 20. The apparatus of claim 18 wherein said first 2 and second relationships are also determined for a time 3 variation of said image.

    1
21. 21. The apparatus of claim 11 wherein said 2 apparatus comprises an imaging system.

    1
22. 22. The apparatus of claim 21 wherein said imaging 2 system is one of a scanner and a digital camera.

    1
23. 23. The apparatus of claim 11 wherein said 2 converter provide.s said first level of accuracy by improving 3 the signal to noise ratio.