A Σ∆ Fractional-N Closed-Loop Modulator for DECT Dan D.S. Hermansen1,2 , Emil F. Buskgaard1,2 , Poul Olesen1 , Finn Andreasen1 , and Torben Larsen2 RTX Telecom1 and RISC division2 , Center for TeleInFrastructure (CTIF), Aalborg University, Denmark E-mail: {ddh,efb,pol,fa}@rtx.dk, tl@kom.aau.dk Abstract Traditionally, modulators for DECT are implemented using open-loop modulation. However, open-loop modulators are too slow for 100% slot utilization. This causes the well known DECT blind-slot problem. One solution to this is to employ closed-loop modulation in stead. This paper presents a closed-loop modulator solution for DECT. The modulator is based on a combination of Σ∆-techniques and use of a fractional-N PLL. The technique allows for digital phase and frequency modulation without use of mixers or D/A converters in the modulation path. Simulation results show a working modulator that is superior to state-of-the-art DECT transmitters using open-loop modulation. An internal reference frequency of 10 − 30 MHz gives an output PSD that meets DECT specification requirements on unwanted emissions due to modulation. The RMS phase error for a reference frequency of 31.104 MHz is as low as 1.06◦ and the peak phase error is 2.90◦ . The blind-slot problem is solved using the closed-loop modulator as the lock time is 1 − 3 µs which, compared to the guard space requirements of below 50 µs, gives a large room for implementation imperfections. 1. Introduction The Digital Enhanced Cordless Telecommunications (DECT) system is an open standard for cordless telecommunications in private homes and office environments [1]. From an RF perspective the key specifications for DECT are as listed in Table 1. Table 1: Key DECT parameters. Modulation GMSK Data rate 1152 [kbps] Frequency band 1880–1900 [MHz] Bandwidth 1.152 [MHz] No. of channels 10 Access scheme TDMA Channel spacing 1.728 [MHz] Nominal TX Power (NTP) 24 [dBm] For DECT, the bill of materials (BoM) is a paramount business parameter. To remain competitive industry is therefore continuously trying to decrease their BoM. For the RF part, modulators have traditionally been implemented using openloop modulation, as illustrated in Figure 1, as a means to reduce component cost. However, due to design tradeoffs related to the division ratios, open-loop modulators are generally too slow for 100% slot utilization. Input data f ref :R PD Gaussian filter Loop filter f rf :N Figure 1: Traditional open-loop modulator. To solve this, every second slot is skipped to allow for the slow PLL to settle. This causes the well known blind-slot problem experienced in DECT. For use in private homes the blind-slot issue poses no practical limitation but for use in larger office environments it is a problem. One solution to this is to employ closed-loop modulation. This paper presents a closed-loop modulator solution that combines the use of Σ∆ techniques with the use of a fractional-N PLL. The modulator topology, illustrated in Figure 2, is based on a fractional-N PLL where the divider ratio is controlled by the output from a Σ∆-modulator [2]. f ref PD Loop filter f rf 8-bit divider Input data N Gaussian filter SD modulator Figure 2: The architectural structure of the closed-loop Σ∆ Fractional-N modulator [2]. In traditional fractional-N based modulators the data directly controls the division ratio. This results in periodicy in the dithering pattern which produces spurious outputs as well as low frequency phase noise. Through the use of a Σ∆modulator this periodicy is mitigated and the resulting noise performance is improved [2]. For DECT there is no requirement on phase error but if a comparison to GSM is made an RMS phase error below 5◦ is required. The aim of most GSM designers is values below 2◦ to reach competitive performance levels. Based on these numbers the design aim for the presented modulator is an RMS phase error below 3◦ and a peak phase error below 10◦ . 2. Modulator Functionality The input data to the modulator is a ±1 bit stream with a bit rate of 1152 kbps. To increase resolution this input data stream needs to be oversampled. Since the phase detector reference signal already is available this is used as sampling frequency as Figure 2 illustrates. This means that for a 20.763 MHz reference frequency each bit consists of 18 samples. In a practical implementation it is preferable to have the reference frequency as low as possible as this lowers the calculation requirements in the digital components. This lowers the power consumption and allows for cheaper implementations. After oversampling the bit sequence passes through a filter with a Gaussian shaped impulse response. The filter shapes each bit and introduces controlled inter-symbol interference. The resulting samples are represented as floating-point numbers between -1 and 1. The divider is chosen as the input to the loop for the modulated bit stream. For this to work the modulated signal needs to be adapted. For the modulator to reach the wanted RF channel and the nominal peak frequency deviation of 288 kHz the modulated signal is offset-scaled. If the modulator is to transmit on channel 9 (1880 MHz) the average input to the Σ∆-modulator needs to be 1880 MHz divided by the reference frequency. For a 20.736 MHz reference frequency this equals 90.664 [-]. The peak deviation is ±1 at the output of the shaping filter. This span also needs to be scaled to a frequency deviation of ±288 kHz. Again, for a 20.736 MHz reference frequency the output of the filter has to be scaled to 288 kHz divided by 20.736 MHz which equals ±0.00139 [-]. The divider can only divide by integer values and the floating point division values therefore needs to be converted to integers. This conversion introduces noise but by using a Σ∆-modulator as converter the resulting noise is shaped. By this relocation to higher frequencies the Σ∆modulation enables the loop filter to attenuate the conversion noise. Furthermore, due to the randomness of the output signal the Σ∆-modulation removes the fractional spurs seen in traditional fractional-N synthesis. Inside the loop the PLL is assumed to be in lock and stable. This idea of modulating the signal inside the loop by altering the division ratios according to the bit sequence is the key point of the architecture in Figure 2. The divider counts the number of clock cycles from the VCO and sends an output pulse when the count reaches the division ratio. At the same time the divider picks a new division ratio from the Σ∆-modulator for the next up count. This way the output phase of the divider is a ‘division rate’ times slower than the VCO output phase. The phase detector (PD) output is a measure of the phase difference (or phase error) between the divider output phase and the reference clock phase. Due to the digital nature of the modulator topology the phase error changes in steps. The PD output is subsequently lowpass filtered in the loop filter to attenuate all but the corresponding slowly varying DC-value. For the topology in Figure 2 to function the modulation has to pass through the loop and a certain minimum bandwidth is obviously required. This forms a trade-off between attenuating loop noise and creating sufficient modulation bandwidth. The LF signal at the output of the loop filter is used to control the VCO. The VCO has a free running center frequency from which it can deviate up or down according to its positive input DC value. So when a phase error is detected the DC value changes and the VCO reacts by increasing or decreasing the output frequency. The phase information from the VCO is looped back to the divider. 3. Simulation Method and Environment To verify and analyze the performance of the modulator a number of simulations are conducted using M ATLAB . To be able to include non-ideal effects in the different system blocks a modular structure strictly following the block diagram of Figure 2 is chosen. Implementing a basic linear VCO model is straight-forward but the model is also fairly inaccurate. To increase the accuracy of the simulation results the VCO model is refined based on measurements of VCO I/O linearity. The divider is implemented as a simple counter while the Σ∆-modulator is implemented as a 4th order MASH-type converter [3]. The Σ∆-modulator converts the floating-point division ratios from the shaping filter to 8bit integers. The implemented PD model is based on a phase frequency detector (PFD) [3]. Unlike a basic phase detector the PFD can not only lock onto the phase of a signal but also the frequency. The loop filter is modeled by a standard M ATLAB generated Butterworth filter with adjustable order and bandwidth. The simulation sampling rate is 10 to 15 times the reference frequency. To avoid loss of high frequency information in the PFD the principle of area conservation is used [4]. Further, to make the process of simulating the modulator as easy as possible a simulation tool is developed. The simulation tool contains an interface for altering reference frequency, loop filter order and bandwidth, loop gain and many other system parameters. Furthermore, simulation parameters such as sample rate and time duration of the simulation can be set. An evaluation tool is also included. This allows for a noise performance analysis using the PSD of the signal. It is also possible to analyze the phase error and to inspect the curve shape of the time, phase and frequency representations of both input and output signals. All of these signals are presented as graphs and RMS and peak phase error are written in numbers as well. 4. Results Based on the developed simulation environment and set-up a number of simulations are conducted. To evaluate the performance of the modulator two key parameters are considered. These performance metrics are the resulting RMS phase error and the adjacent-channel emission spectra. The simulations have been conducted using all possible combinations of reference frequency (10.368 MHz, 20.736 MHz, 31.104 MHz), loop filter cut-off frequency (1000 kHz, 1200 kHz, 1600 kHz), and loop filter order (2nd , 3rd , 4th ). To evaluate the modulators performance over the entire DECT bandwidth the simulations are repeated for each channel. Figures 3 to 5 show the average RMS phase noise performance for all channels while Figures 6 to 8 show the worstcase emission performance for all channels. meet the phase noise design target with the best performance resulting from a third-order filter with a cut-off frequency of 1600 kHz. 10 1000 kHz 1200 kHz 1600 kHz 9 13 12 11 RMS phase error [deg] RMS phase error [deg] 8 1000 kHz 1200 kHz 1600 kHz 10 9 6 5 4 8 3 7 2 6 1 5 4 2 3 Loop filter order [−] 4 Figure 3: Mean phase error versus filter order for various filter bandwidths for a reference frequency of 10.368 MHz. From Figure 3 it is found that the combination of a low filter order and a high cut-off frequency results in the worst noise performance. Increasing the filter order improves on the noise performance in all cases but for the lowest cut-off frequency. In all cases the phase noise performance fails to meet the target value below 3◦ . This result indicates that a reference frequency of 10.368 MHz does not provide sufficient room for noise shaping of the dither signal. 7 1000 kHz 1200 kHz 1600 kHz 6 RMS phase error [deg] 7 5 4 3 2 1 2 3 Loop filter order [−] 4 Figure 4: Mean phase error versus filter order for various filter bandwidths for a reference frequency of 20.736 MHz. Comparing Figures 3 and 4 it appears that the tendency from Figure 3 is reversed for a 20.736 MHz reference frequency. Here the best noise performance is achieved for low filter orders. This indicates that the noise shaping is better and that signal degradation results from loop filter group-delay distortion. The increased noise level for the second-order, 1600 kHz cut-off filter shows that the noise from the Σ∆modulator still has some impact. Six of the combinations 2 3 Loop filter order [−] 4 Figure 5: Mean phase error versus filter order for various filter bandwidths for a reference frequency of 31.104 MHz. For a 31.104 MHz reference frequency the noise from the Σ∆-modulator is insignificant as Figure 5 illustrates. In this case the group-delay distortion from the loop filter is the most significant source to signal degradation. The tendency is clearly that an increased filter order and a reduced cut-off frequency degrades phase noise performance of the modulator. From Figures 3 to 5 it is found that there exists a tradeoff between the reference frequency and the sharpness of the loop filter. At a low reference frequency the noise from the Σ∆-modulator is dominant and sharp filtering increases performance. At higher reference frequencies the noise from the Σ∆-modulator becomes less significant and loop filter group-delay effects become more important. At the highest reference frequency group-delay is dominant and relaxed filtering provides for best performance in this case. The last performance parameter to evaluate is the adjacentchannel emission. When this performance metric is considered the performance of the modulator is found to depend slightly on the RF channel setting. To simplify result-graphs only worst-case performance is presented. For the low reference frequency the emission results are presented in Figure 6 together with the emission envelope requirement. From this it is clearly seen that for a second-order loop filter the modulator is not able to comply with emission requirements. The third-order filter also fails for a cut-off frequency of 1600 kHz. With the reference frequency increased to 20.736 MHz, as illustrated in Figure 7, the emission performance is seen to increase for almost all filter combinations. Only a single combination, the second-order 1600 kHz cut-off filter, fails to meet requirements. For the second-order, 1200 kHz cutoff filter the performance surplus is very small and there is almost no room for implementation margin. When the reference frequency is increased to 31.104 MHz only the secondorder, 1600 kHz cut-off filter fails to meet requirements with sufficient margin. For all remaining combinations more than 10dB of margin is achieved. 40 Adjacent channel emmisions [dBm] 20 0 −20 −40 40 Adjacent channel emmisions [dBm] 1000 kHz, 2nd order 1200 kHz, 2nd order 1600 kHz, 2nd order 1000 kHz, 3rd order 1200 kHz, 3rd order 1600 kHz, 3rd order 1000 kHz, 4th order 1200 kHz, 4th order 1600 kHz, 4th order 20 0 −20 −40 −60 −60 −80 −80 1 It is clear that the best emission performance is achieved for high-order filters with a low cut-off frequency. Since the loop filter effectively serves as an output filter for the RF channel this result comes as no surprise. 40 1000 kHz, 2nd order 1200 kHz, 2nd order 1600 kHz, 2nd order 1000 kHz, 3rd order 1200 kHz, 3rd order 1600 kHz, 3rd order 1000 kHz, 4th order 1200 kHz, 4th order 1600 kHz, 4th order 20 0 1 2 3 4 Adjacent channel number [−] Figure 6: Worst-case emission levels for a 10.368 MHz reference frequency. NTP is 24 dBm, channel bandwidth is 1.152 MHz, and channel spacing is 1.728 MHz. Adjacent channel emmisions [dBm] 1000 kHz, 2nd order 1200 kHz, 2nd order 1600 kHz, 2nd order 1000 kHz, 3rd order 1200 kHz, 3rd order 1600 kHz, 3rd order 1000 kHz, 4th order 1200 kHz, 4th order 1600 kHz, 4th order −20 2 3 4 Adjacent channel number [−] Figure 8: Worst-case emission levels for a 31.104 MHz reference frequency. NTP is 24 dBm, channel bandwidth is 1.152 MHz, and channel spacing is 1.728 MHz. frequency of 31.104 MHz. However, for cost and power consumption reasons a low sampling frequency is wanted. Unfortunately, at low sampling frequencies the conversion noise from the Σ∆-modulator becomes significant and phase noise performance is degrade. Furthermore, at the lower reference frequencies the Σ∆ noise shaping is insufficient and adjacent-channel emission performance is also degrade. A third-order loop filter with a 1600 kHz bandwidth operated in conjunction with a 20.736 MHz reference frequency is found to be a suitable tradeoff. The best channel RMS phase error for this combination is 1.60◦ and a peak phase error of 4.62◦. For this set-up the lowest margin on the emission due to modulation is 8 dB which leaves sufficient room for implementation margin. With lock-times of around 1 − 3 µs the modulator topology is found to be well suited for implementing DECT transmitters. −40 6. References −60 −80 1 2 3 4 Adjacent channel number [−] [1] ETSI, Digital Enhanced Cordless Telecommunications (DECT); Common Interface (CI); Part 2: Physical Layer (PHL), ch. 4-6. ETSI, etsi en 300 175-2 v1.7.1 ed., 2003-07. Figure 7: Worst-case emission levels for a 20.736 MHz reference frequency. NTP is 24 dBm, channel bandwidth is 1.152 MHz, and channel spacing is 1.728 MHz. [2] M. H. Perrott, Techniques for High Data Rate Modulation and Low Power Operation of Fractional-N Frequency Synthesizers, ch. 2,8,9,11. Massachusetts Institute of Technology, 1997-09. 5. Conclusions [3] M. Kozak, Oversampled delta-sigma modulators : analysis, applications, and novel topologies, ch. 2,3,5. Boston, Massachusets : Kluwer Academic Publishers, 2003. A closed-loop modulator based on Σ∆ noise shaping is presented. Based on simulations it is found that a large loop filter bandwidth and a low filter order results in low phase error. Further, it is found that a low bandwidth and a high order reduces spurious emissions. The best phase noise performance, 1.06◦ RMS and 2.9◦ peak, is obtained for a reference [4] M. H. Perrott, Fast and Accurate Behavioral Simulation of Fractional-N Frequency Synthesizers and other PLL/DLL Circuits. MIT/DAC, 2002-06.