Field of the Invention This invention relates to analog-to-digital converters, and particularly (though not exclusively) to such converters as may be used in integrated circuits for use in wireless communications.

Background of the Invention In the field of this invention it is known that, in practice, converting an analog signal to the digital domain with a resolution equal to or higher than 12 bits requires using a sigma-delta modulator. Not using an oversampled converter, as in a pipeline converter for example, leads to stringent capacitor matching requirement, thus limiting the resolution to 10-12 bits; it is well known that sigma-delta converters allow this limitation to be overcome. However, the oversampling requirement inherent to sigma-delta converters has historically limited the signal bandwidth. There has been a trend in the recent past to increase bandwidth for application in the wireless communication area, such as for GSM (Global System for Mobile communications) or CDMA (Code Division Multiple Access) bandwidth. However, converting a wider bandwidth with a resolution of 12 bits or more requires new solutions. For example, converting the WCDMA (Wideband CDMA) baseband signal means

converting a bandwidth of nearly 2MHz. Several approaches could be considered, each of them having its own drawbacks.

Starting with Nyquist rate converters, as mentioned above, pipeline converters lead to stringent capacitor matching requirement which usually limits the resolution below 12 bits. Additionally, handling a 2MHz bandwidth would require amplifiers with a very high unity gain bandwidth, up to 500MHz.

A second alternative of a Nyquist rate converter is a timely-interleaved multipath pipeline converter. This alternative would lower the conversion rate of each pipeline converter but in general suffers from a fixed pattern noise resulting in input offset and gain matching errors among the parallel pipeline converters. The capacitor matching requirement would also remain an issue for resolutions above 12 bits.

Considering an oversampled converter, the appropriate shaping of the quantisation noise either requires increasing the order of the sigma-delta modulation or necessitates a very high sampling frequency. Stability or matching constraints do not make it possible to extend the order to a very high extent and, overall, the sampling frequency could have to be increased far beyond 100MHz. This would typically require extra circuits such as a PLL (Phase Locked Loop) to generate the sampling clock. Moreover, for many implementations a strict

requirement might be put on the phase noise of the PLL in order to achieve 12-bit resolution in the converter.

Another possible alternative is a Hadamard modulator architecture, which would lower the sampling rate but would require many parallel paths and hence would occupy a large integrated circuit silicon area.

A need therefore exists for a sigma-delta analog-to- digital converter arrangement and method wherein the abovementioned disadvantage (s) may be alleviated.

Statement of Invention In accordance with a first aspect of the present invention there is provided a sigma-delta analog-to- digital converter arrangement as claimed in claim 1.

In accordance with a second aspect of the present invention there is provided a method for sigma-delta analog-to-digital conversion as claimed in claim 17.

Brief Description of the Drawings One parallel translating sigma-delta analog-to-digital converter arrangement and method incorporating the present invention will now be described, by way of example only, with reference to the accompanying drawings, in which:

FIG. 1 is a block schematic diagram illustrating a parallel translating sigma-delta analog-to-digital converter arrangement incorporating the present invention.

FIG. 2 illustrates a switching sequence whose use in the sigma-delta analog-to-digital converter arrangement of FIG. 1 would produce significant third and fifth harmonics.

FIG. 3 illustrates the spectrum associated with the switching sequence of FIG. 3.

FIG. 4 illustrates switching sequences which could be used for frequency translating in the sigma-delta analog-to-digital converter arrangement of FIG. 1.

FIG. 5 illustrates the spectrum associated with the switching sequences of FIG. 4.

FIG. 6 illustrates a timing diagram representation of the switching sequences of FIG. 4.

FIG. 7 illustrates a possible embodiment for a sigma-delta modulator differential input stage used in the sigma-delta analog-to-digital converter arrangement of FIG. 1 with the switching sequences of FIG. 4.

FIG. 8 illustrates the spectrum of an example of an input signal applied to the sigma-delta analog-to- digital converter arrangement of FIG. 1.

FIG. 9 illustrates the spectrum of an example of a sum of switching signals in the sigma-delta analog- to-digital converter arrangement of FIG. 1.

FIG. 10-12 illustrate the spectra of examples of intermediate signals at various points along the processing flow of the sigma-delta analog-to-digital converter arrangement of FIG. 1.

FIG. 13 illustrates the spectrum of an example of the signal at the output of the sigma-delta analog- to-digital converter arrangement of FIG. 1.

Description of Preferred Embodiment Briefly stated, a preferred embodiment of the present invention converts analog signals into digital signals by combining two identical low-pass sigma-delta converters in such a way that the signal bandwidth is extended by a 2-fold factor.

Referring firstly to FIG. 1, an analog-to-digital converter (ADC) arrangement 100 receives an analog input signal 10, and delivers a digital output signal 90. The ADC arrangement 100 is split into two parallel paths:

the first path comprising translating sigma- modulator block 21, decimation filter and down- sampling block 31, DC notch filter block 41 and mixer block 51; and the second path comprising translating sigma- modulator block 22, decimation filter and down- sampling block 32, DC notch filter block 42 and mixer block 52.

As will be explained in more detail below, in the first and second paths the split signals are subjected to phase shifts nominally in quadrature and are processed in reduced bandwidth frequency bands, to produce the summed digital output signal 90 with a frequency bandwidth greater than the reduced frequency band.

Blocks 21 and 22 are low-pass sigma-delta modulators.

Low-pass sigma-delta modulation is a well known function to quantize a signal and sample it at a much higher rate than the Nyquist rate such that most of the quantisation noise is pushed towards high frequencies. Thus, the SNR (Signal-to-Noise Ratio) in the low pass region can be made high, making the modulator suitable for processing baseband signals. The SNR, and therefore the resolution, and the signal bandwidth where this SNR is achieved can be increased by increasing the order of the sigma-delta modulator (SDM) and the sampling frequency. However, this principle can not be applied beyond certain limits.

Therefore, in the arrangement 100 the signal bandwidth is doubled by having a sigma-delta modulator only processing half of the bandwidth of the signal to be converted with

an appropriate SNR. For this purpose, the analog input signal 10 is first frequency-translated at the input of each SDM 21 and 22.

In a discrete-time SDM, the sampling operation occurs in the input stage of the SDM, usually by sampling the analog input signal onto a sampling capacitor by means of a switched-capacitor arrangement. It is well known that a switching operation mixes an input signal with the switching signal controlling the switches. By adequately choosing the switching signals, it is therefore possible to produce a frequency translation of the spectrum of the input signal. However, in the ADC arrangement 100 this frequency shift must be in the order of half the signal bandwidth to be converted. Considering as an example a WCDMA signal, the bandwidth to be converted is in the order of 2MHz and the frequency shift needs to be in the order of 1MHz. A switching signal is a digital signal and, therefore, may exhibit a high level of harmonics, especially when considering that the switching signal needs to be built relative to the sampling frequency.

Taking as an example a sampling rate of 30.72MHz, which is an 8-fold multiple of the 3. 84MHz chip rate of a WCDMA receiver, a basic switching signal could alternate sixteen states to 1 and sixteen states to-1. Here, the state definition relates to an analog input signal applied in a well-known differential mode between a positive node and a negative node. A state of 1 on the switching signal means that the differential analog input signal is sampled in the SDM input stage in a

straightforward manner while a state of-1 means that the differential analog input signal is sampled in a cross- coupled manner, the positive input node being sampled and applied to the negative section of the differential SDM structure and the negative input node being sampled and applied to the positive section of the differential SDM structure. One period of such a basic switching scheme is illustrated in FIG. 2. The frequency spectrum of this switching signal is shown in FIG. 3, from which it may be observed that the spectrum contains many high level harmonics. The closest ones, namely the third harmonic (H3) and the fifth harmonic (H5) would mix with interferers that could be present in the analog input signal (e. g. , in an adjacent channel and an alternate channel of wireless communication system) and would be folded in the same frequency band as the wanted signal present in the analog input signal. This drawback could be avoided by drastically filtering these interferers prior to the ADC operation. However, such filtering would be expensive.

The present invention allows for a simpler solution: Instead of using a single sampling structure, several sampling structures may be combined together in order to remove some harmonics from the spectrum resulting from their combination. A well known arrangement meeting this property uses Walsh functions as switching sequences.

However, a drawback of using Walsh functions is the need for non-integer weighting factors which would have to turn into non-round values of sampling capacitors in the switched-capacitor input stage of the SDM. This would

increase the sensitivity of the overall ADC performance to the capacitor mismatch.

The ADC arrangement 100 relies on using a combination of switching sequences which are delayed against each other such that the third and fifth harmonics are removed or at least very significantly attenuated in the frequency spectrum of their combination. FIG. 4 shows two sets of switching sequences having this property. The first set comprising CI1 to CI4 is intended for use in the first SDM, 21, of FIG. 1. The second set comprising CQ1 to CQ4 is intended for use in the second SDM, 22, of FIG. 1.

Summing together the sequences CI1 to CI4 or CQ1 to CQ4 produces the spectrum shown in FIG. 5, from which it may be observed that this spectrum does not exhibit any significant third or fifth harmonic contribution.

The nearest harmonic is the seventh, thus significantly relaxing the requirements of the analog filter needed in front of the ADC. In practice, such a filter is required anyway for anti-aliasing purpose with regard to the sampling frequency used in the ADC. Applying the switching sequences shown in FIG. 4 allows the analog filter requirement not to be increased with regard to the usual anti-aliasing constraint.

It will be understood that alternatively to using switching sequences having unit coefficients as shown in FIG. 4, switching sequences having binary weighted coefficients in combinations of the FIG. 4 switching

sequences (e. g. , +2, 0,-2 or +4, +2, 0,-2,-4) may be used.

FIG. 7 shows an example of the way the switching sequences of FIG. 4 may be applied and combined into a switched-capacitor input stage of a SDM. Such a classical differential stage would just use two equally- sized sampling capacitors switched by means of switches controlled by several non-overlapping or delayed clock phases, namely X 2 and old in FIG. 7. In the present embodiment of the invention, eight equally sized capacitors may be used instead of two and the switches may be controlled by a logic combination of the clock phases and the control sequences CI1 to CI4 in the first SDM, 21, or CQ1 to CQ4 in 22. As this solution uses equally sized capacitors, this minimises the sensitivity to capacitor mismatch. The switching sequences have been translated into control signals in the timing diagram of FIG. 6.

It may be noted that the periodicity of the switching sequences determines that the fundamental is equal to the sampling frequency divided by 30 in this example. Thus, this would make the fundamental frequency equal to 30.72MHz/30 = 1.024MHz, which is approximately half of the signal bandwidth to be converted. An alternative solution for implementing the combination of the switching sequences could consist in adding the four sequences all together and using just binary-weighted capacitors with fewer switches. This would also keep the sensitivity to capacitor mismatch low.

Another alternative could be an intermediate solution where the switching sequences are added by pairs and applied, by means of a reduced number of switches compared to the first embodiment, to double-size capacitors.

In the diagram of FIG. 1, both SDM's are identical.

However, the analog input signal is translated in a different way due to the phase shift between switching signals CI1 to CI4 on one side and CQ1 to CQ4 on another side. This translating operation will be compensated further along the processing flow by blocks 51 and 52 which are mixers having quadrature mixing signals 54 and 56 (and which provide a reverse process of frequency translation and phase shifting to that which occurred in the translating sigma-delta modulators 21 and 22 respectively). Therefore, the switching sequences applied to SDM's 21 and 22, or at least their combination or their overall effect should be in quadrature too.

Actually they just need to almost be in quadrature. It may be observed in FIG. 4 and 6 that CI1 to CI4 are not perfectly in quadrature with CQ1 to CQ4 respectively.

This allows the sampling frequency to be kept low while still having the appropriate delay between each switching signal. In FIG. 4 the quadrature error is n/30. This error may be compensated by adjusting the phase of the mixing signal 56 by means of the parameter 9.

The order and the structure of the SDM's 21 and 22 has to be chosen according to the SNR performance of the ADC.

Just considering the SNR of one SDM, the relevant spectrum portion for evaluating the SNR performance is half of the overall signal bandwidth that the ADC 100 is supposed to handle. Therefore, the SNR performance of one SDM should be the same as the target performance of the overall ADC because it handles approximately half of the total signal power and the half of the total noise power. In the diagram of the parallel translating ADC 100 of FIG. 1 the blocks 31 and 32 next to the SDM's 21 and 22 perform the decimation filtering and down-sampling of the signals 24 and 26 respectively, coming from blocks 21 and 22 respectively. Decimation filtering is a well known operation in a low-pass sigma-delta ADC. It allows removal of the quantisation and electronic noise between the top of the signal bandwidth and half of the sampling frequency used in the SDM. This makes it possible to re- sample the signal, after the decimation filtering, at a much lower frequency, usually close to the Nyquist rate with regard to the signal bandwidth. The decimation filtering may be handled by a single digital filter or by a cascade of several digital filters of various known types such as SINC (sin (x) lx) filters, FIR (Finite Impulse Response) filters, half-band filters, etc. In the parallel translating ADC arrangement 100 the transition band of the decimation filtering, between the passband and the stopband, needs to be sharp and very close to the end of the signal bandwidth handled by a SDM, i. e. , half the total signal bandwidth of the overall ADC. In the above example the transition band needs to be slightly above 1 MHz. The selectivity of filtering beyond this limit is needed to ensure the proper behaviour of the

overall ADC, 100. After the decimation filtering and the re-sampling of the digital signal, signals 34 and 36 pass through blocks 41 and 42 respectively in order to remove the DC offset produced in the SDM. If these blocks are omitted, the DC offset will be shifted in the next block, 51 or 52, at a non-DC frequency, thus producing an unwanted tone in the signal bandwidth. Omitting these blocks is therefore possible only if the DC offset is small enough to avoid degrading the SNR performance of the overall ADC. Blocks 41 and 42, called DC notch'in FIG. 1, perform high pass filtering signals of the signals 34 and 36 respectively, if needed. The cut-off frequency of this filter may be chosen as low as needed to avoid degrading the wanted signal.

The next processing stage comprises blocks 51 and 52.

These blocks perform a mixing operation by multiplying the signal coming out of the DC notch filters, 44 and 46, by single-tone signals in quadrature, 54 and 56. These signals (as already explained) may have a slight phase adjustment to compensate for a quadrature error made on the switching signals at the input of the SDM's. The frequency of these single-tone signals needs to be equal to the fundamental of the switching signals. As both types of signals are digital, it is obvious that this relationship can easily be guaranteed. Moreover, it will be easily understood by a person of ordinary skill in this art that the single-tone mixing signals can be generated from the SDM sampling signal with the required frequency and phase accuracy by means of well-known

arrangements, that need not be described in further detail.

The final stage in the ADC comprises an adder 80 to combine signals 64 and 66 coming out of the mixers 51 and 52 respectively. The adder 80 delivers a digital signal 90 with an SNR characteristic approximately equal to the SNR performance of each SDM, 21 or 22, while the SNR characteristic is available over the double of the bandwidth of each SDM. This allows sigma-delta modulation to be applied to double bandwidth signals without having to double the sampling frequency or to increase the complexity of the SDM.

FIG. 8 to 13 illustrate the signal processing for the example of sinusoidal signals spread in the WCDMA bandwidth. FIG. 8 shows a two-tone signal applied at the ADC input 10, with a frequency component at 600kHz and another one at 1800kHz. FIG. 9 shows the spectrum of the sum of the switching signals used in the input stage of each SDM, either made by summing CI1 to CI4 or CQ1 to CQ4 respectively. The fundamental, which is going to produce the frequency shift in this stage, is at 1024kHz.

Third and fifth harmonics are visible in this spectrum because some capacitor mismatch has been introduced for the purpose of illustration in the SDM input stage.

FIG. 10 shows the spectrum of the signal at the output of the SDM stage, blocks 21 and 22. It should be noted that this is an amplitude spectrum; therefore, the phase difference is not visible in such a spectrum and FIG. 10 may serve as an illustrating plot for both signals 24 and

26. Two low frequency tones may be observed, at 1024kHz-600kHz = 424kHz and at 1800kHz-1024kHz = 776kHz. This illustrates the fact that the ADC input tones, which were spread in a 2MHz bandwidth, now have contributions in a lMHz bandwidth where the quantisation noise is the lowest. As the quantisation noise results from a sixth order sigma-delta noise shaping in this example, a much higher quantisation noise level may be observed beyond 1 MHz. Two extra tones may also be noticed at 1024kHz + 600kHz = 1624kHz and 1024kHz + 1800kHz = 2824kHz.

FIG. 11 shows the spectrum of signals 34 and 36 at the output of the decimation filtering and down-sampling blocks 31 and 32. At this point the quantisation noise beyond 1200kHz and the high frequency tones have been significantly attenuated. FIG. 12 shows the spectrum of signals 64 and 66 at the output of mixers 51 and 52. The mixing operation has restored components at 600kHz and 1800kHz but unwanted components are visible at 1024kHz-776kHz = 248kHz and 1024kHz + 424kHz = 1448kHz.

Finally, FIG. 13 shows the spectrum of signal 90 at the output of the ADC, after adder 80. Re-combining signals 64 and 66 has allowed the unwanted tones at 248kHz and 1448kHz to be significantly reduced while the original tones at 600kHz and 1800kHz have been increased by 6dB.

The unwanted tones could be fully eliminated if capacitor matching were better. However, for illustration purposes this imperfection has been exaggerated in this example.

In fact, a resolution of the ADC beyond 12 bits could

still afford a capacitor mismatch in the order of one percent, which is easily achievable.

It will be understood that the parallel translating sigma-delta analog-to-digital converter described above provides the following advantages: It allows the signal bandwidth to be doubled for almost the same resolution without the need for doubling the sampling frequency. This reduces the power consumption compared to a solution with a double clock speed, making this converter particularly suitable for use in wireless telephones.

Both the embedded sigma-delta modulators and the digital filters and mixers lend themselves for being adjusted to other communication standards. This makes this converter suitable for multi-mode or multi-standard wireless telephones.

In the above-described converter, with respect to such a wireless telephone application, the rejection of an interferer in an adjacent channel and an alternate channel is related to the ability of the translating sigma-delta modulator to suppress the third and fifth harmonics respectively. This suppression has been achieved with the particular switching sequences of the translating sigma-delta modulators.

It will be understood that the converter will typically be fabricated in an integrated circuit (not shown).

It will be further appreciated that other alternatives to the embodiment of the invention described above will be apparent to a person of ordinary skill in the art.