TITLE: COMBINATION D/A CONVERTER AND FIR FILTER
UTILIZING ACTIVE CURRENT DIVISION AND METHOD
Specification Cross-Reference to Related Patent Applications
This patent application is a continuation-in-part of patent application Serial No. 08/389,362, filed 02/16/95, which is pending.
Background of the Invention
1. Field of the Invention. The present invention relates to a digital-to-analog FIR filter. More specifically, the present invention relates to a digital-to-analog filter which uses active current division steering techniques.
2. Brief Description of the Related Technology.
Previous methods for digital-to-analog (D/A) signal conversion and reconstruction filtering involve several methods. The actual D/A conversion process has been accomplished using methods such as a single current source and sink, or by dumping charge from a switched capacitor that has been charged to either a positive or negative reference voltage. Reconstruction filtering has been accomplished by using a combination of active and/or passive classical filtering techniques, such as continuous time active filters with resistors and capacitors, continuous time passive filters utilizing resistors, capacitors, and inductors, or switched capacitor filter techniques.
A method of filtering which combines a D/A converter with a reconstruction filter has recently become known in the art. Oversampled D/A converters generally include the following signal processing blocks: (1) an interpolator filter, or series of filters, which raises the sample rate of the incoming digital signal to a higher sample rate, (2) a digital sigma-delta processor (or noise shaper) which lowers the number of
bits representing the signal by shaping the quantization noise in a way that places most of it at higher frequencies, (3) a D/A converter which converts the output of the noise shaper into an analog signal, and (4) an analog low pass filter which removes, or substantially lowers, the noise that was placed at higher frequencies by the noise shaper.
In all sigma-delta D/A converters, there exists a need to filter the high frequency noise inherent to this method of conversion. It is common for a digital noise shaper (digital sigma-delta modulator) to have as its output a single bit. The single bit digital output signal is then converted to an analog signal using switched capacitor techniques or switched current source techniques. Once this conversion is made, filtering of the high frequency noise is then accomplished through a variety of means.
As illustrated in Fig. 1, a semi-digital reconstruction filter typically uses a tapped delay line, or shift register, to control a plurality of devices each of which has an associated gain factor. The outputs of the plurality of devices are then summed together to form a single output of the filter. In some cases, individual current sources are employed as the plurality of devices. The amount of current in each current source is designed such that a desired FIR filter response is achieved. The output of each current source is then provided, or steered, to a current summing node (IOUT) or to an alternative current summing node (IOUT*), depending on the logic state of the control bit (BN) at the delay line tap associated with each current source. The currents at one or both of the current summing nodes are then converted to a voltage using standard current to voltage conversion techniques. Additional filtering may then be employed to remove extremely high frequency noise.
In other cases, the plurality of devices in the semi-digital filter FIR coefficients are represented as charge stored on a plurality of capacitors. The charge on each capacitor can then be summed by employing a switched capacitor summing amplifier. Once again, additional filtering may be employed to remove any extremely high frequency noise.
In another semi-digital filter scheme, the FIR coefficients are represented as a current value through a plurality of resistors. Each resistor is selectively connected to a voltage reference depending on the state of the
individual control bit from the delay line tap associated with each individual resistor. The current is then summed and converted to a voltage by means of resistive feedback around an operational amplifier (op amp). As in the previous methods discussed, additional filtering may be employed to remove any high frequency noise.
Previous current steering semi-digital FIR filters utilize current paths having linear resistors, or resistive elements such as FETs or CMOS transmission gates biased in the linear region. This results in relatively low output impedance of the resistive elements of the filter. Any offset on inputs to an operational amplifier (op amp) connected to the current paths in the prior filter circuits may cause an error term. Since the current through each path is dependent on the resistance of the resistive element in each path, the state of the switches and the op amp offset, the FIR coefficients for the filter, as determined by the current in each path, are a function of the op amp offset. A need exists to minimize, or eliminate, this offset distortion term.
Summary of the Invention The method and apparatus described herein is one which uses a single current reference and an active current divider network which includes a plurality of parallel current paths. The current through each path is actively steered by transistors to a current summing node, or an alternative current summing node, depending on the logic level of a control bit at a delay line tap associated with each path. A shift register which shifts the single bit, output signal from a digital sigma-delta modulator, forms a delay line. The shift register includes a series of flip-flops. The output of each flip flop provides the input of the next flip flop in the series. All flip-flops are clocked at the same rate. The output of each flip flop also provides the control bit associated with each current path. That is, for an individual path, if the control bit is a logic 1, the current through the path is actively steered to the current summing node. If the control bit is a logic 0, the current is actively steered to the alternative current summing node. Since a tapped delay line and a summed output(s) consisting of
weighted values (with weight determined by conductance value of each path) of the intermediate points in the tapped delay line are provided, a Finite Impulse Response (FIR) filter exists. Such a filter may be designed using techniques available to those skilled in the art. Any type of filter such as bandpass, bandreject, high pass, and low pass may be built using this technique. The present invention is directed to, a low "pass filter which is desirable to remove high frequency noise generated by a sigma-delta data conversion process.
Transistors biased in the saturation region are utilized as active current steering elements in each current path to actively steer current. This results in the elements in each current path having a relatively high output impedance, which minimizes the effect of any op amp offset. This causes the FIR filter signal response to be more reliable than when passive resistive elements are utilized.
Brief Description of the Drawings
Figure 1 schematically illustrates a prior art embodiment of a semi- digital FIR filter utilizing a plurality of current sources;
Figure 2 schematically illustrates a prior art semi-digital FIR filter which uses a single current source and passive current steering techniques; Figure 3 schematically illustrates a prior art embodiment which removes an effective DC offset using a current source;
Figure 4 schematically illustrates an embodiment of the present invention which utilizes differential currents and differential voltages;
Figures 5a-c schematically illustrate various prior art embodiments to implement the resistive paths or RO and Rl associated with the switches of
Figs. 2 and 3;
Figure 6 schematically illustrates a semi-digital FIR filter of the present invention which uses a single current source and active current steering techniques; and Figure 7 schematically illustrates an embodiment of the present invention which removes an effective DC offset using a current source.
Detailed Description of the Preferred Embodiment
The present invention utilizes a single reference current, IREF, an active current division network, a method for steering the output of each individual transistor element in the current steering network, a current summing node, an alternative current summing node, and a method to convert the current in the current summing node to a voltage.
Figure 1 illustrates a prior art implementation of a semi-digital FIR filter 50 using a plurality of current sources 52 and 53. The 1-bit output from a sigma-delta modulator 16 is input to shift register 14 as input signal 12. Sigma-delta modulator 16 preferably includes an interpolation circuit and a noise shaping circuit. Various interpolation circuits known in the art may be utilized, however, the preferred digital interpolation circuit is described in application Serial No. 08/333,399, filed November 2, 1994, entitled "Digital Interpolation Circuit for A digital-to-analog Converter Circuit", assigned to the common assignee of the present invention. Likewise, various digital noise shaping circuits may be utilized, however, the preferred noise shaper circuit is described in application Serial No. 08/333,386, filed November 2, 1994, entitled "Digital Noise shaper Circuit", assigned to the common assignee of the present invention. The current flows through the non-inverted current switches (B0, B,) to non-inverted current summing node 62 and through inverted current switches (B0*, B,*) to inverted current summing node 58. Switches B0 and B0* are controlled by the logic level of output tap B0 of shift register 14. Switches Bx and Bj* are controlled by the logic level of output tap B,. If output tap B0 is a logic 0, inverted current switch B0* is closed and non-inverted switch B0 is open, causing current to flow from current source 52 to inverted current summing node 58. If output tap B0 is a logic 1, non-inverted current switch B0 is closed and inverted current switch B0* is open, causing current to flow from current source 52 to non-inverted current summing node 62. Non- inverted current switch B, and inverted current switch Bj* would function in a similar manner, causing current to flow from current source 53 to non- inverted current summing node 62 or inverted current summing node 58, depending on the logic value of shift register output tap B
Figure 2 illustrates a prior art semi-digital FIR filter current steering circuit 10. Figure 2 depicts a current steering circuit 10 which includes two control bits (B0 and B,) and their logical inverses (B0* and B^) which are output from shift register 14 as output tap B0 and B^ respectively. These control bits are used to control whether the current in an individual resistive path 21, 23 is steered to non-inverted current summing node 62 (IOUT), or to the alternative current summing node, inverted current summing node 58 (IOUT*). In order for the individual currents through the resistive paths 21, 23 to remain constant, the current summing nodes 62, 58 (IOUT and IOUT*) must be held at identical voltages. For the example shown in Fig. 2, it will be assumed that summing nodes 62 and 58 (IOUT and IOUT*) are at zero volts. Thus, the current through resistive path 21, represented by resistive element RO, is represented by the following equation: lo = IREF [R,/ (Ro-r-R,)] Likewise, the current through resistive path 23, represented by resistive element R-, is given by the following equation:
I, = IREF [ly (Ro+R,)] Therefore, the current in non-inverted current summing node 62 (IOUT) can be represented by the following equation: IOUT(k) = I0 [x(k)j + II x [x(k-l)] where x(k) is the digital input signal 12, illustrated in Fig. 1, where input signal 12 (x(k)) is output from sigma-delta modulator 16, and where IOUT(k) represents the output current in the non-inverted current summing node 62, and I0 and I- are the currents through the two resistive paths 21, 23, for R-, and R-, respectively.
The standard equation for an FIR filter is given by: y (k) = ao [x(k)] + a, [x(k-D] + ... an [x(k-n)] Since the equation for non-inverted current IOUT(k) is of the same form as for y(k), the structure illustrated in Fig. 1 is an FIR filter. The equation for non-inverted current IOUT(k) can also be written in the following way by substituting I0 and Ij as their equivalent functions of current source IREF, resistive elements R„ and Rj, as given in the above equations by: IOUT(k)=[IREF[R1/(R0+R1)]] x(k) + [IREF [R-ΛRo+Rj)]] x(k-l)
Table 1 provides the summed currents IOUT and IOUT* for all possibilities of switches B0 and B, in Fig. 2.
Since, in Fig. 1, input signal 12 (x(k)) can take on only a value of logic 0 or logic 1, current summing nodes IOUT and IOUT* can only be equal to zero or positive values. In fact, both current summing nodes IOUT and IOUT* may take on values from zero to the value of current source IREF (Fig. 2).
Thus, the structure illustrated in Fig. 2 adds an effective DC offset with a value of IREF/2. This can easily be removed by subtracting a fixed amount of current, IREF/2, from current summing nodes IOUT and IOUT*, as shown in Fig. 3. Table 2 illustrates that both current summing nodes, IOUT and IOUT*, may take on values from -IREF/2 to IREF/2.
In fact for each combination of switches B0 and B15 the value of inverted current summing node 58, IOUT*, is equal to negative the value of non- inverted current summing node 62, -[IOUT]. Thus, the inverted and non- inverted currents are differential in nature. This differential current embodiment is used to remove any even-ordered distortion which may occur. Figure 4 illustrates a method of converting current summing nodes IOUT and IOUT* from differential currents to differential voltages via operational amplifier circuits 20 and 22. Each op amp circuit, 20 and 22, includes an op amp 25 and a feedback resistor 30. If desired, the differential voltages in this embodiment may be converted to a single-ended voltage 37 using series input resistor RA, voltage divider resistor network Re and 1 and feedback resistor Rβ together with op amp 27, as shown in Fig. 4. As stated previously, additional filtering of extremely high frequency noise can then be accomplished by connecting a capacitor CF in parallel to feedback resistor 30 in Fig. 4.
Several prior art techniques may be used to implement the resistive paths 21 and 23, via resistive elements R0 and Rl and associated switches B0, B0*, B-, and B,* of Figs. 2 and 3. Resistive elements Ro and ^ may be resistors, as shown in Figs. 2 and 3. As shown in Fig. 5a, an explicit resistor
R, and a pair of switches B, and B,* are utilized to implement resistive paths 25 and 27. The resistor R, may be realized as a poly resistor, a diffused resistor, a thin film resistor, or by any of the standard methods of realizing resistors. The switches B, and B,* may be realized by employing CMOS transmission gates Tj and T2, or single MOSFETS, to act as switches. In the
technique of Fig. 5a, the resistance of an individual resistive path 25 or 27 would be the sum of the explicit resistor, R„ plus the effective "on" resistance of the switch B, or B,*. Care must be taken to make the "on" resistance of T. and T2 negligible with respect to R„ or, alternatively, the "on" resistance of the switch B, or B,* may also be ratioed by the same ratio which is used to calculate R,. However, like the prior art techniques shown in Fig. 1, discussed above, any error in effective static switch "on" resistance will only change the effective FIR coefficient for the filter and will not affect the linearity of the filter. Another technique that may be used to implement resistive paths 25 and 27 is shown in Fig. 5b, is to realize R-o or Rj in Figs. 2 and 3 as the effective "on" resistance of a CMOS transmission gate T,, or single MOSFET transistor biased in the triode (linear or resistive) region when turned on. In Fig. 5b, Tj represents the effective resistance for R- or R The switches B, and B,* are likewise realized as CMOS transmission gates T2 and T3, or as single MOSFET transistors. The switches B, and B,* are designed such that the effective "on" resistance of gate T2 is identical to the effective "on" resistance of gate T3. Thus, the effective resistance of the resistive paths 25 or 27 is the sum of the "on" resistance of gate T, plus the "on" resistance of gate T2 or T3. Like the technique described above for Fig. 5a, the "on" resistance of gate T2 and T3 must either be made negligible with respect to the "on" resistance of gate T^ or they must be ratioed by the same ratio which is used to calculate the on resistance of gate Tr Once again, any error in effective static switch "on" resistance will only change the effective FIR filter coefficient and will not affect the linearity of the analog signal at current summing nodes 62 and 58, IOUT and IOUT*.
An alternative prior technique that may be used to implement resistive paths 25 and 27, shown in Fig. 5c. Resistive elements R„ and R, and switches B0, B0*, Bj, and B,* are implemented as individual CMOS transmission gates TA and TB, or as single MOSFET transistors. In this case, the "on" resistance of gate TA is designed to be equivalent to the "on" resistance of gate TB. Thus, the effective resistance of the resistive paths 25 or 27 is the "on" resistance of gates TA or TB. Like the above methods shown in Figs. 5a or 5b, any error
in effective static "on" resistance will only change the effective FIR filter coefficient (and therefore the frequency response of the filter) and not the linearity of the analog signal which results from using this technique.
The prime reason the current steering structure of Figs. 2 and 3 is linear is because the effective current through each resistive path 21 or 23 is not dependent on the current through any of the other resistive path. Thus, ideally, the current through a resistive path 21 or 23 should not depend on the 1-bit digital input signal 12, x(k), of Fig. 1. This is true for the ideal cases of the three techniques described above and illustrated in Figs. 5a-c. In practice, however, this may not be completely true.
If, in Fig. 5c, due to random manufacturing resistance mismatches, the "on" resistance of gate TA is not equivalent to the "on" resistance of gate TB, assuming, for example, that TA and TBB are the same size, nominally, then current IA passing through switch TA when shift register output tap B, is a logic 1 would not be equivalent to current IB through gate TB when output tap
B, is a logic 0. This means current I, through a resistive path 25 or 27, rather than being constant, is dependent on the input signal 12, x(k). This concept also applies to the embodiments illustrated in Figs. 5a and 5b. Since the FIR filter structure as a whole is built as a current divider, if the current in one resistive path 25 or 27 changes, the amount of current through all paths 25 and 27 must change in order for the total current to remain equal to reference current IREF. Thus, there is potential for harmonic distortion to result. One of the first assumptions is that the value of current summing nodes IOUT and IOUT* must be held at the same potential. In practice, this may not be possible due to random op amp offset voltages. Fortunately, to the extent the harmonic distortion is even-ordered, using differential non- inverted and inverted currents, IOUT and IOUT*, and converting them to differential voltages as shown in Fig. 4, will reduce the total harmonic distortion. Figure 6 illustrates a FIR filter circuit 50 and method of the present invention for actively steering current through a plurality of current paths. The single current source, IREF is biased by signal VBIASl. IREF is divided into smaller branch currents by means of MOS transistors T0-T2, which are
biased in the saturation region, to actively steer current through current branches, or paths, 31, 33 and 35, respectively. Transistors T0-T2 are biased by signal VBIAS2 in the saturation region, which results in the transistors having a relatively high output impedance. The amount of current in each branch of the FIR filter circuit 50 is determined by the effective W/L ratio of each individual transistor, T0-T2, with respect to the sum of all the W/L ratios. Since each transistor has the same gate voltage and the same drain voltage, those transistors with larger W/L ratios will conduct proportionately larger current than those with smaller W/L ratios. The W/L ratio of each transistor is sized to implement effective coefficients for the FIR filter circuit 50.
The current through each branch is defined by the following equation: I, = IREF * (W/L), / (W/L , In the above equation, IREF is the single master current, (W/L), is the W/L ratio for each transistor and (W/L) ^ is the sum of all W/L ratios for all transistors. Thus, a semidigital FIR filter having (n+ 1) taps implemented with this technique would produce an output current described by the following equation:
IOUT = Z0 * IREF + Zl * IREF * Z"1 +Z2 * IREF * Z2 + ... + Zn * IREF *z n Where Z, is (W/L), / (W/L)toul As illustrated in Fig. 7, an active DC offset compensation current steering filter circuit 100, having differential current outputs IOUT and IOUT*, includes DC offset current sinks IREF/2 connected to IOUT and IOUT* to reduce the effect of the DC offset, for the reasons discussed previously in regard to Figs. 2 and 3. The differential current-to-voltage conversion circuit 29 of Fig. 4 is connected to the active current steering FIR filter circuit 50 of Fig. 6 or circuit 100 of Fig. 7, such that the IOUT lines of Figs. 4 and 6 are connected, and the IOUT* lines of Figs. 4 and 6 are connected, respectively. With the relatively high output impedance of transistors T0-T2 in Figs. 6 and 7, the effect of any offset at the inputs of op amps 25 of Fig. 4 are minimized.
The switches, B0 and B0*, etc. of Figs. 6 and 7 may be implemented using CMOS transmission gates, or using single MOSFET transistors biased in the linear region.
The present invention, therefore, is well adapted to carry out the objects and attain the ends and advantages mentioned herein as well as other ends and advantages made apparent from the disclosure. While preferred embodiments of the invention have been described for the purpose of disclosure, numerous changes and modifications to those embodiments described herein will be readily apparent to those skilled in the art and are encompassed within the spirit of the invention and the scope of the following claims.
What is claimed is: