In order to increase the dynamical range of the HSRL, new photon counting electronics were designed (Figure 8). The photon counting data system has three counters and each counter has 2 buffers, one for each polarization. Counting rates have been tested up to 350 MHz, but counters are designed to operate up to 1 GHz. The 100 ns bin width enables a 15 m range resolution. The 4 kHz laser repetition rate enables the use of 2333 range bins, which can be used to probe the atmosphere up to 35 km. The limiting factor of the data system is the photomultipliers (EMI 9863B/100), which allow counting only up to 15 MHz without significant pile-up effects. The PMT's are pile-up and afterpulse tested and both corrections are included to the data processing. The corrections are discussed later in this chapter.
Computer control of the system is made by using Intel 80960CA embedded processor. The user interface is realized by using a Sun Sparc Station II. In addition to the data taking and the communication with the Sun, the system processor is used to control HSRL operation including pressure controlling of the etalons, timing of the Pockels cell voltage, controlling of the seedlaser temperature, controlling of the system shutters and relays, and controlling of the WFOV aperture.
Most of the HSRL-operations are computer controllable from a menu interface. System performance can be followed from real time displays. The real time Range Time Indicator (RTI) displays for the raw and/or inverted data are used to display the incoming data so that altitude, variability, thickness and depolarization of the cloud layer can be observed when the clouds move over the lidar site. Similar displays for the cloud optical properties (optical depth, aerosol backscatter cross section, phase function) can be shown.
Figure 8: The HSRL data acquisition system.
The new counter design has improved the background light measurement accuracy of the HSRL. Instead of interrupting the data taking every 3.5--7 min for background and calibration checks (as was made with the system described by Grund ), the new HSRL is capable of measuring the background light simultaneously with the data. The maximum resolution of the data system is increased from 256 to 4096 range bins per channel. In the HSRL, the amount of background light in the measurement is calculated from the measured profile by using the upper altitudes of the lidar profile. The signal together with the background light is saved and therefore different kind of background corrections can be studied. Currently, the measured HSRL profiles are background corrected by assuming the signal above 33 km to be background. By averaging the upper range bins together, the average amount of background in the measurement is calculated. The disadvantage of the current background correction method is that some of the signal can still be originated from Rayleigh scattering leading into an overestimate of the amount of background light for altitudes 33 km or higher. A development of a background correction method that removes the Rayleigh scattering contribution from the background is currently in progress.
Photomultiplier afterpulsing can be seen as multiple output counts for a single input photon . The afterpulsing could be caused by positive electron production in the dynode chain. The afterpulse pulses are time delayed compared to the signal from incident photon and the time delay is proportional to the path length for electron propagation. The signal originated from a short light pulse and measured with the HSRL can be presented in following form
= signal measured with the HSRL
S(t') = signal incident on the receiver
= signal responce function, that contains the laser pulse width, datasystem bin width and afterpulsing of the photomultiplier
The afterpulse probability distributions were measured by illuminating the photomultiplier photocathode with a short duration laser light pulse. Pulse counting rates less than one count per 100 ns long data bin per pulse were used. A large number of laser pulses were averaged (1 hour average). The probability distributions were calculated from the background corrected data. The amount of background light was calculated from the end of data record and the average value was subtracted from data. The measured afterpulse probability distributions of two photomultiplier tubes are presented in Figure 9. The afterpulse probability distribution was found to be different for each tube and therefore a separate correction for each tube had to be implemented to the signal analysis program. With a known afterpulse probability distribution function, the afterpulse contribution can be removed from the measured signal by using deconvolution. The effect of afterpulse correction to the measured signal is presented in Figure 10.
Figure 9: The temporal distributions of two EMI 9863B/100 photomultiplier tubes. The temporal distribution of the measured photons is a convolution of the laser pulse width, bin width of the data system, and the afterpulsing of the photomultiplier. The laser pulse is located in the first bins and the afterpulse can be seen as a decaying tail. The afterpulse distribution of the PMT1 shows better afterpulse behavior than the PMT2.
Figure 10: The effect of afterpulse correction to the measured aerosol signal in the case of thick water cloud (at 5.6 km) with an ice crystal precipitation at 5.1 km. The afterpulsing increases the aerosol signal at the end of the cloud, and therefore the geometrical thickness of the clouds is underestimated. The effect of afterpulsing to the molecular signal is similar, and thus a slight overestimate of the optical depth of the cloud is made, if the afterpulse correction is not performed.
The pulse pile-up is seen when all time separated photons are not distinguishable and they overlap. This causes a nonlinear response. In order to be able to separate between counts, a minimum separation time between two pulses is required for them to be distinguished. This time is called resolving time (or dead time). Since the time interval distribution of photons that reach the photomultiplier follow a Poisson distribution, the probability that a pulse overlaps with another pulse inside a certain time interval is given by the following formula 
= resolving time
R = average count rate
In order to obtain the resolving time (dead time) of the photomultipliers, the photomultiplier responses to different count rates were tested. First, light level is set to level which is high enough to generate a pile-up. Then the photomultiplier response to different light levels is measured by changing the light intensity with neutral density filters and recording the signal change. In order to be able to define the resolving time of the photomultiplier, the photon counting was modeled. The pulses coming from the photomultiplier were treated as Poisson distributed signals. The resolving time that produced the best fit between simulated and measured results was used for the pile-up correction. All photomultipliers were found to have 13 ns resolving time. The pile-up correction has to be performed before the afterpulse correction or any other corrections, because the pile-up effect is nonlinear. The pile-up is affected by the signal strength, signal background and afterpulsing and therefore, data has to contain the information from all these factors when the correction for pile-up is performed.
In order to make the pile-up correction for data, the following equation was used
N = measured counts
= actual counts
= resolving time
R = average count rate =
This equation is solved iteratively for , so that
1. = N
2. New , where dt = bin length [s]
3. Step 2 is repeated until converges
The effect of pile-up correction to the measured signal is presented in Figure 11.
Figure 11: The effect of pile-up correction to the measured aerosol signal in the case of thick water cloud (at 5.6 km) with an ice crystal precipitation at 5.1 km. The calibration fiber signals are shown as spikes at altitudes 0 km. The maximum number of photons measured for the water cloud is 0.68 photons/100 ns bin and after pile-up correction 0.75 photons/100ns bin are observed. Even with this low count rates the effect of pile-up is 10% and therefore, the pile-up correction is very important when strong returns from clouds are studied.