Atomic absorption spectrometry with a photodiode array spectrometer

Enhancement of wavelength measurement accuracy using a linear photodiode array as a detector for inductively coupled plasma atomic emission spectromet...
0 downloads 0 Views 990KB Size
Anal. Chem. 1980, 52, 2133-2140

2133

Atomic Absorption Spectrometry with a Photodiode Array Spectrometer E. G. Codding' Department of Chemistry, University of Calgary, Calgary, Alberta, Canada T2N 1N4

J. D. Ingle, Jr. Department of Chemistry, Oregon State Universw, Cowallis, Oregon 9733 1

A. J. Stratton' Department of Chemistry, Kent State University, Kent, Ohio 44240

An evaluation of a self-scanned linear photodiode array as a multkhannel detector for flame atomic absorption is presented. The noise sources contributing to the overall measurement precision were determined at array integration times of 1 and 5 s. The photodiode array spectrometer was found to be electronic readout noise limited under ail experimental conditions for the analytes studied (Ag, Cr, Cu, Mg, Mn, Zn), with the exception of Mg which displayed slgnlficant anaiyte absorption flicker noise. Analytical figures of merit are presented for the species studied and are seen to be somewhat inferior to literature data but superior to previously reported values for a slmiiar instrumental system. The utillty of the photodlode array as a multielement detector Is demonstrated and dlscussed.

Recent reports in t h e literature have demonstrated and discussed the application of solid-state, multichannel detector systems, such as the linear silicon photodiode array (PDA) (1, 2) and the two-dimensional silicon vidicon (3-6), as multielement detectors for flame atomic absorption measurements. Although the primary use of these multichannel detector systems has been for multielement analysis, other potential advantages have been discussed in the above reports. These include the use of nonspecifically absorbed lines which are near the analyte line for simultaneous background correction and of several analyte absorption lines t o improve measurement precision and to extend the linear concentration range. T h e important points t o consider in assessing the potential application of a measurement system are the measurement precision which may be expected and the identification of those parameters which limit that precision. These have not been thoroughly addressed. Chuang (2) has considered several algorithms for the extraction of analytical information from flame atomic absorption spectral data acquired with a photodiode array. He has also presented experimental sensitivity and detection limit data for several analyte species. However, no attempt was made t o identify and evaluate the parameters limiting the measurement precision. In several recent reports (7-10)the factors affecting the relative precision of flame atomic absorption (AA) measurements with a conventional AA spectrophotometer (a spectrophotometer having a single exit slit and utilizing a photomultiplier (PMT) detector) have been discussed for a variety of analyte species. It was shown that, in general, signal shot Present address: Varian Associates, Palo Alto, CA. 0003-2700/80/0352-2133$01 .OO/O

noise, hollow cathode lamp (HCL) flicker noise, or flame transmission noise limit the measurement precision a t low absorbance (0-0.1A ) , analyte absorption flicker noise dominates a t moderate absorbance (0.1-1.0 A ) , and eventually background emission flicker and analyte emission flicker noise become limiting at high absorbance (>1.0 A ) . The purpose of this report is t o quantitatively evaluate a PDA as a multichannel detector for AA measurements and to identify those experimental/instrumental parameters which limit the measurement precision.

EXPERIMENTAL SECTION The PDA-AA spectrometer used for all measurements consists of a Varian AA-6 burner/nebulizer with a 10-cm single-slotburner head and a laboratory constructed gas control system, a GCA McPherson EU-700 monochromator, and a Reticon RL512C/ 17 PDA with a RC 100/102 readout board. Westinghouse and Cathodean HCLs were used as sources and were powered by a Heath Schumberger EU-703-62 power supply. Detailed descriptions regarding the use and general operating characteristics of PDAs have appeared in the literature (11, 12) and will not be repeated here. A block diagram of the monochromator/data system is shown in Figure 1. The PDA consists of a linear array of 512 photodiodes, or pixels, which are 12.7 pm wide, 432 fim high, and on 25.4 pm centers. Thus, the active area of the array is 13 mm long. The array is cooled to approximately -10 "C with a thermoelectric cooler to reduce thermally generated dark current. The Reticon readout board was used as received with the exception that three additional 9316 binary counters were added to allow greater flexibility in the selection of integration times. The video output from the PDA was digitized to 12 bit resolution with a successive approximation analog-to-digital converter (BurrBrown ADC80AG-12) which was in turn interfaced to a PDP llV/03 computer via a DRVll parallel line interface for data acquisition and manipulation. The PDA was clocked at 31.5 kHz which was limited by the data acquisition capability of the computer. Thus, 16.2 ms were required to readout a single scan of all 512 pixels in the PDA. All programming was carried out under RT-11 in FORTRAN IV with the exception of three MACRO subroutines to effect the data acquisition and display functions. The block diagram of the optical/burner/monochromator configuration used for the flame atomic absorption measurements is shown in Figure 2. Radiation from the HCL was focused a t the center of the bumer head and refocused at the monochromator entrance slit by two f/3.0 supracil lenses. For multielement analysis the radiation from a second HCL was combined with the first by placing a polka dot beamsplitter in the optical path and the second HCL at right angles to the optical path as shown in Figure 2. This method of combining the output of HCLs is acceptable for perhaps as many as three lamps; however, as more beamsplitters are introduced the radiant throughput becomes too low for practical applications (13). The gas flow rates and burner 0 1980 American Chemical Society

2134

ANALYTICAL CHEMISTRY, VOL. 52, NO. 13, NOVEMBER 1980

Table I. Experimentally Measured Signals Required for Precision Analysis

c

I I

f

Ed, ud: 100% T voltage and rms noise; flame on, HCL on, blank aspirating E o t , u o t : 0% T voltage and rms noise; flame on, HCL off, blank aspirating Est, ust: sample voltage and rms noise; flame on, HCL on, analyte aspirating Eet, uet: analyte emission voltage and rms noise; flame on, HCL off, most concentrated analyte solution aspirating E*:, u,;: lamp voltage and rms noise; flame off, HCL on E,, , uot): electronic signal and rms noise; flame off, HCL off A3 in the Appendix to calculate the relative precision ( c r ~ / A ) .

Figure 1. Block diagram of the photodiode array spectrometer and data system. BEAM SPLITTER 1

SLOT

Block diagram of the optical system for atomic absorption measurements.

Figure 2.

position were optimized for each analyte species studied. The GCA McPherson monochromator used in this work had its exit folding mirror removed to accommodate the PDA detector assembly as described elsewhere (11,12). The combination of the RL512C/17 PDA with this monochromator and grating allows a spectral range of about 28 nm to be simultaneously monitored. AA measurements were made for the following elements where the analytial wavelength in nanometers and hollow cathode lamp in milliamps, respectively, are given in parentheses: Ag (328.1, 3), Cu (324.7, 4), Cr (357.9, 6), Mg (285.2, 2), Mn (279.5, 15), Zn (213.9,6). All measurements were made with an air/C2H2flame utilizing a 0.25-nm monochromator band-pass (100-pmslit width). Single element HCLs were used for all elements except Mn where a Cr-Ni-Mn-Fe lamp was used. To take full advantage of the dynamic range of the PDA, it is necessary for the analytical emission line from the HCL to nearly saturate the PDA for the selected integration time. This can be accomplished by increasing either the HCL current or the PDA integration time. However, it was felt to be desirable Lot to operate the lamps at excessive currents because of the recognized deleterious effects on the analytical performance (13). A survey of several HCLs confirmed that near saturation levels could be obtained with the lamp operating at, or near, recommended currents with a 5.053-s PDA integration time. To illustrate the effects of not having near saturation PDA output levels, we made measurements with a 1.090-sintegration time without changing the HCL current. Thus the PDA output signal should be reduced by a factor of 4.636, the ratio of the two integration times. For the magnesium studies a maximum integration time of 2.147 s was used because of the high spectral output from the Mg lamp at the recommended operating current. For brevity, these integration times will hereafter be referred to as 1, 2, and 5 s, respectively. The six experimental parameters which must be determined to allow evaluation of the experimental and theoretical relative precision (uA/A) are summarized in Table I. These parameters were measured for each of the analyte species studied at two integration times. Replicate measurements of each parameter were obtained by acquiring, and storing individually, 32 consecutive scans of the PDA. The mean and standard deviation were obtained for each corresponding point in each of the 32 spectra. These data were then used along with eq Al, A2, and

RESULTS AND DISCUSSION General Considerations. The use of a multichannel detector, such as a silicon PDA provides the unique advantage of simultaneous measurement of spectral data over a wavelength range which is determined by the linear dispersion of the monochromator and the length of the active surface of the array. However, when compared to a conventional single-element AA spectrophotometer utilizing a PMT, certain disadvantages result. The dynamic range of a PDA is limited by electronic readout noise introduced during readout and the onset of saturation at long integration times. The saturation level for a typical PDA such as employed here is 2 X lo7 electrons. The total readout noise depends on the noise characteristics of the PDA (reset noise of the pixels and dark current noise) and of the readout circuitry (amplifier noise) (14). Typically a root-mean-square readout noise of 5-30 X 103 electrons can be obtained with the Reticon readout boards while values of 1-2 x lo3 electrons have been obtained with laboratory-constructed and commercial readout circuitry (14). Thus, dividing the saturation level by this range of rootmean-square readout noise yields dynamic ranges of about 700-20000 compared to over 1 X lo6 for a PMT. In the case of multielement analysis utilizing a multielement HCL, the most intense element line intensity can be adjusted by taking advantage of the PDA integration time and HCL current to utilize the full dynamic range of the PDA, and thus realize a typical relative readout resolution about 0.1% for one integration period with the systems used here. However, if a second analyte line is only one-tenth as intense, ita relative readout resolution will be 10 times worse, about 1% , assuming the flame transmission and detector quantum efficiency are equal at both wavelengths. For this reason it would be more desirable t o use single-element HCLs and combine their outputs with beamsplitters ( 2 , 1 3 ) ,thus allowing the output intensity of each lamp to be individually adjusted to near the PDA full scale value. The complexity of combining several HCL outputs in this manner coupled with reduced intensities would limit the number of elements for stimultaneous analysis to a few. In a conventional AA spectrophotometer, the HCL output is modulated, or chopped, and lock-in amplifier type detection is used. Thus, the measurement system does not respond to dc flame background or analyte emission signals and lowfrequency l / f flicker noise in these signals is greatly reduced. With a PDA detector system, an unmodulated HCL is used t o provide the maximum radiation intensity during the integration period. Since the detector integrates all of the signal, the maximum usable integration time is determined by the time required for the sum of dark current, HCL radiation, analyte emission, and flame background emission to saturate the PDA. Thus unlike a conventional AA, a separate 0% transmittance ( r ) measurement must be made in addition to the reference and sample measurements as a base-line cor-

ANALYTICAL CHEMISTRY, VOL. 52, NO. 13, NOVEMBER 1980

rection for dark and background emission signals. If analyte emission is significant, its contribution to the absorption signal must be measured, along with dark current and flame background emission, and it must be corrected for at each analyte concentration. Thus, the dc nature of the PDA measurement system makes it more susceptible to l / f noise in the analyte and flame background emission signals than a conventional AA-PMT measurement system. A modulated PDA-AA system could be implemented by synchronizing the PDA integration time to the HCL on time. Since typical HCL modulation frequencies in AA are about 100 Hz, a 5-ms integration time would have to be employed. With the HCL currents employed, the lamp signal would be about of full scale and essentially equal to the rootmean-square readout noise. One thousand 5-ms integrations could be summed to yield a total integration time of 5 s, but the relative standard deviation in the HCL signal due to readout noise would be (1 x 103)1/*greater than with a continuous 5-s integration with one readout. Thus with modulated or unmodulated HCL signals, it is always best to achieve the desired total measurement time by keeping the maximum HCL signal near saturation and summing an appropriate number of integrations or readout cycles. This will keep the relative readout noise to a minimum since the absolute readout noise increases in proportion to the square root of the number of scans. The expected signal-to-noise ratio (S/N) differences when a PDA is used in place of a P M T have been briefly discussed (7). Equation A3 in the Appendix describes the manner in which the relative precision in absorbance (uA/A) depends upon absorbance ( A ) or T and the experimental variables (7-10). T h e terms in the brackets account for all the noise sources and are respectively the signal shot noise (KE,(1 T ' ) )analyte , emission noise (crJrnGT)', HCL flicker noise (2(ir4Jz), flame transmission noise (2(i,.$JZ),0% T noise (2(1 + T z- T l ) ( ~ ~ / r n Gand ) ~ )analyte , absorption noise ((2).The 0% T noise, eq A4, is seen to be due to background emission noise (abe) and 0% T'noise (aot'). T h e 0% T'noise is the residual noise observed when the flame and HCL are off and is due to dark current noise and electronic readout noise. With the modifications shown in the Appendix, eq A3 applies to a PDA-AA systems. Note the effect of having to make a separate 0 % T measurement is incorporated into the equations. Some further comparisons can be made between P M T and PDA detector systems, assuming that the two AA spectrophotometers are identical with the exception of detectors. The following variables are functions of the detector: i,, ue, and uot (see Appendix for definitions). The photocathodic current, i,, photoelectrons, or electron-hole pairs produced per unit time will differ because the quantum efficiency and size of the detectors differ. For comparison purposes, a 1P28 P M T has a useful wavelength range of approximately 200-650 nm. Its quantum efficiency drops off precipitously near the limits of this range and has a value of about 20% at 300 nm. A PDA generally displays a smooth and gradual decrease in quantum efficiency from its maximum a t approximately 600 nm down to 300 nm and a slight increase from 300 to 200 nm (15). From 200 to 900 nm, the quantum efficiency of a PDA is above 30% and greater than 50% from about 400 to 800 nm and around 200 nm. Thus the PDA has a quantum efficiency advantage of a factor of 2 or greater a t most wavelengths compared to a typical P M T . A typical entrance slit height for a conventional AA spectrophotometer would be 5-10 mm, although a 2-4 mm slit height may be used without signal loss due to the diameter of the HCL image formed at the entrance slit (16). Since the diameter of a typical P M T photocathode is greater than this,

+

2135

IC

Figure 3. The 324.7-nm Cu emission line from hollow cathode lamp.

the total entrance slit image will be measured by the P M T . PDAs are available with a maximum pixel height of 2.5 mm, which would be comparable to the HCL image, although the device employed in this study has a pixel height of only 0.43 mm resulting in only about 1C-20% of the available light being measured. Typically about a 100-pm slit width is used in AA measurements. T h e active width of each pixel is effectively 25 pm since the actual diode (p-type bar) is 13 pm wide and charge generated by light on the 13 pm wide n-type material between p regions will divide between the adjacent diodes (16). Thus, each pixel intercepts about 25% of the width of a HCL image in the focal plane compared to a PMT. Hence, due to quantum efficiency and size considerations, i, may be typically 2-20 times less in PDA systems for the same integration time. Since the HCL image will overlap a number of diodes, the signals from several diodes can be summed to effectively increase i, although the absolute readout noise will be in proportion to the square root of the number of diodes summed. Because of the smaller size of the detector element the analyte and flame background emission signals will be smaller with a PDA by the same fraction that the lamp photocathodic signal is reduced. The effect on ue and am is more difficult to predict. The absolute value of the fundamental shot noise in these emission signals will decrease as the square root of the emission signal decreases. However, the unmodulated signal processing in the PDA-AA spectrometer make it susceptible to l / f noise in the emission signals which could ultimately make the emission noises larger relative to the reference signal (E,) and worsen the cr~/A,particularly at higher absorbances.

DATA EXTRACTION Figure 3 shows a short section, 32 points, of a copper HCL emission spectrum in the region of the Cu 324.7-nm line which has been acquired with the PDA spectrometer and plotted on an X-Y recorder. It is clear that the extraction of analytical information from a sampled spectral line such as the one shown is not straightforward. If care is taken to ensure that the central sampled value corresponds to the peak of the emission line, as is the case in Figure 3, the peak value could be used as a measure of the intensity. However, using only the peak value does not take advantage of the intensity information contained in the other sampled points across the emission line profile. Further, adjusting the monochromator to place a single pixel at the peak of the emission line becomes tedious and only under fortuitous circumstances would a second analyte emission line also have a single pixel simultaneously positioned a t the peak of the line. As indicated

2136

ANALYTICAL CHEMISTRY, VOL. 52, NO. 13, NOVEMBER 1980

Table 11. Output Voltage for Seven Pixels Centered at the Peak of the Cu 324.7-nmEmission Line for 1-and 5-s Integration Times 0.2 ppm Cu point no. E,, art E,, a,t E,, ast 1-Integration Time 1 2 3 4a 5 6 7

1 2 3 5 6 7 a

0.009 0.009 0.007 0.007 0,008 0.008 0.008

0.414 0.628 0.839 0.904 0.822 0.618 0.402

0.007 0.007 0.008 0.007 0.008 0.007 0.008

5-s Integration Time 1.965 0.008 0.299 0.008 3.078 0.008 0.289 0.007 4.093 0.009 0.342 0.008 4.488 0.008 0.284 0.008 4.062 0.009 0.292 0.007 3.097 0.008 0.272 0.009 1.972 0.008 0.274 0.007

1.886 2.945 3.912 4.293 3.890 2.967 1.899

0.008 0.009 0.008 0.009 0.007 0.008 0.008

0.434 0.659 0.879 0.951 0.864 0.652 0.417

0.007 0.008 0.009 0.008 0.007 0.008 0.008

0.082 0.085 0.069 0.076 0.068 0.074 0.007

Table 111. Results of Using Only Peak Value and Summing 3, 5 , and 7 Values across the Cu Emission for 0.2 Ppm Cu Analyte Solution at 1-and 5-s Integration Time values parameter

5

3

7

1-s Integration Time

0.882 0.835 0.75 0.66 0.68 0.024 20.9

2.464 2.335 1.37 1.12 1.27 0.023 14.9

3.643 3.449 1.78 1.16 1.78 0.024 12.3

4.345 4.116 2.07 1.19 2.07 0.024 12.3

5-sIntegration Time 4.204 11.725 17.339 4.009 11.178 16.529 0.79 1.48 1.85 0.92 1.24 1.69 0.79 1.31 1.75 0.021 0.021 0.021 6.2 3.5 3.1

Peak of Cu emission line.

above, the photocathodic area of a P M T is typically much larger than the spectral image formed at the exit slit so that a P M T effectively integrates the spectral information contained in the emission peak. Hence, a similar integration approach seems advisable here. Chuang (2) has evaluated two algorithms for extracting such analytical data: a quartic fit of the seven highest sampled values and subsequent calculation of the peak value, and secondly, the average value of the five highest pixels. While the quartic fit yielded the best precision, it was set aside in favor of the averaging technique because of the uncertain bias introduced by the quartic method. However, it does not seem appropriate to average the intensity values of the pixels since this implies that these points should have the same value, viz., that the emission line should have a flat top rather than be peak shaped as is observed. In this report we evaluate a technique which involves simply summing the intensity values of the sampled points symmetrically about the peak value of the emission line which is akin to performing a digital integration over the emission peak. This is statistically identical with calculating the average value except the sum of the signals is not divided by the number of diodes used. T h e summation method was evaluated by measuring the experimental parameters described above and tabulated in Table I with a 0.2 ppm Cu analyte solution and optimized experimental conditions. The experimental average values and standard deviations for E,,, Eo,, and E,,, at 1- and 5-s integration times for 32 scans of the PDA were obtained. The results for seven pixels, three each side of the peak value, are shown in Table 11. There are two noteworthy observations to be made from Table 11. First, comparing the standard deviations of the measured values for each integration time shows that these values are essentially identical. This suggests that the dominant noise source is independent of the absorbance and is probably electronic readout noise. Second, comparing the standard deviations between the two integration times again shows that they are essentially equal. This implies that the dominant noise source is independent of integration time. These two observations could be accounted for if the dominant noise source was electronic readout noise which would be introduced only upon readout of the PDA. Thus, the relative precision ( u / E )would improve as the integration time was increased since E increases and u remains constant. Quantitative data regarding the noise contributions are presented below.

1

20.703 19.741 2.15 2.03 2.04 0.021 3.1

Table IV. Absorbance, Standard Deviation and Relative Precision Calculated for Seven Individual Pixels across the Cu 324.7-nm Analytical Line, 0.2Ppm Cu Solution Aspirating into Flame 1-sintegration

point no. 1 2 3 4a

5 6 7 a

A

0.026 0.024 0.022 0.024 0.024 0.025 0.020

UAX

100

1.24 0.79 0.67 0.50 0.60 0.80 1.40

5-sintegration

~AIA x

100

A

48.4 33.2 30.1 20.9 25.2 31.5 72.5

0.021 0.021 0.021 0.021 0.020 0.021 0.019

UAX

100

0.29 0.19 0.14 0.13 0.10 0.17 0.31

~ A J AX 100

13.9 8.9 6.7 6.2 5.1 8.2 15.7

Peak of HCL emission line.

A comparison of the measurement precision obtained by using only the peak value and from summing one, two, and three values each side of the peak value is shown in Table 111. E, and E, have been corrected for dark current by subtracting Eo, from Ertand Est as shown in eq A l . Appropriate propagation of error methods have been used throughout. In comparing the results it should be noted that the standard deviations for the 1-s and 5-s integration times for the various sums are essentially identical suggesting again that the dominate noise source is independent of integration time as would be the case if electronic readout noise was limiting. The summed values were used with eq A2 to calculate the experimental relative precision in absorbance (uA/A) as tabulated in Table I11 for the two integration times. The results illustrated in this example for 0.2 ppm Cu are typical for the other elements studied. There is a dramatic improvement in the relative precision in going from using only the peak value to summing three values and a smaller improvement when five values are summed. However when seven values are summed, three each side of the peak, there is either no further improvement or a slight deterioration, in the relative precision. The reasons for this can be seen with the aid of the data presented in Tables 11and IV and Figure 3. Table IV presents A , uA,and uA/A for the data of Table 11. As has been noted for the data of Table 11, the absolute measurement precision is effectively invariant from pixel to pixel. However, as the spectral intensity decreases away from the peak value, the relative measurement precisions for E , and Est decrease since E,, and E,, decrease. This decrease in the relative measure-

ANALYTICAL CHEMISTRY, VOL. 52, NO. 13, NOVEMBER 1980

I ,

01 0

2

4

6

8

I3

I2

4

L

Figure 4. Precision plots for Cu at different integration times: (a) 1.023 s; (b) 5.053 s;(c) theoretical precision plot corresponding to curve b, all noise sources considered.

ment precision results in a corresponding increase in uA, and hence uA/A, with which the absorbance can be determined as shown in Table IV. Thus, when values further away from the peak value are included in the sum, a point is reached where the overall precision no longer improves but rather becomes worse. For the instrumental system being used here, the optimum number of pixels to be summed is five, two each side of the peak value. These results are expected since the absolute readout noise increases with the square root of the number of diodes. Thus when a diode signal is less than 1/2lI2 of the peak diode signal, or uA/A of a diode is greater than 2lI2 of the peak diode, no S I N ratio advantage results from using the signal from that diode. The summation of five values will be used in the remainder of this report.

ANALYSIS OF PRECISION The experimental relative precision plots, (uA/A) X vs. A calculated from eq A2, for the elements tested are shown in Figures 4 and 5 , and the calculated parameters are given in Table V. Figure 4 shows the experimental relative precision plots for copper a t 1- and 5-s integration times. The corresponding data in Table V indicate that the noises (precision) associated with those measurements which were performed directly: 0% T noise, am; 100% T noise, ufi; 100% T'noise, (re' and electronic readout noise, uo(, for a given integration time, are all approximately equal. Further, the relative electronic noise, aot'/Er,is essentially equal to, or greater than, each of the other relative noises. This substantiates the earlier suggestion that the dominant noise source is electronic readout noise. The same observation is made for the other elements tested. The nature of the electronic readout noise is not understood. It originates during the PDA readout process and is presumably due to capacitive crosstalk between the video output line and the various clocking signals within the PDA and also from clocking transients within the RC100/102 readout board (14). T h e PDA and its associated readout electronics, up to the analog-to-digital converter, have been operated from a battery with no improvement in the noise characteristics. A comparison of the noise contributions between the 1-and 5-s integration time data of Table V shows that the relative

2137

2138

ANALYTICAL CHEMISTRY, VOL. 52, NO. 13. NOVEMBER 1980

Table VI. Analytical Figures of Merit for Single-Element Atomic Absorption Determinations element Ag

cr cu Mg Mn Zn

integration time, s 5 1 5 1 5

sensitivity,=ppm this work ChuangC 0.046 0.050 0.183 0.170 0.044 0.043 0.004 0.004 0.045 0.052 0.017 0.016

1 2 1 5 1 5 1

lit.d 0.029

0.15

0.055

0.14

0.040

0.02

0.003

0.05

0.021

0.03

0.009

detection limit,b ppm this work Chuanf lit.d 0.014 0.066 0.069 0.316 0.014 0.066 0.001 0.003 0.011 0.062 0.004 0.019

0.002 0.03

0.005

0.02

0.002

0.006

0.0002

0.01

0.002

0.006

0.001

a Sensitivity defined as ppm to cause 1%absorption of incident light. Detection limit defined as ppm to cause A = 2 0 ~ . Chuang et al., ref 2, p 525. Varian Analytical methods “Cookbook”.

1°T I Ll

f/number than that used for the P M T studies. The higher i, for Zn may be do the higher quantum efficiency of the PDA near 200 nm. Thus, the relative shot noise ( ( , J ~ ) ~ + is ~ /about E,) the same with both detectors and insignificant compared to the readout noise of the PDA. Evaluation of the remaining noise sources identified by Ingle such as source flicker noise, flame transmission noise, and analyte absorption noise is not possible because of the dominance of electronic readout noise. Only for Mg with a 2-s integration time was significant analyte absorption flicker noise observed since ‘T,~ was 1.2 times greater than the electronic readout noise for A = 0.1-0.6.

(a,

a

I

_-

Figure 5. Experimental precision plots: (a) Mg (2.147-s integration time);(b) Cr (5.053-sintegration time);(c)Ag (5.053-sintegration time); (d) Mn (5.053-sintegration time); (e) Zn (5.053-sintegration time).

noises for the 1-s data are approximately 4.6 times greater than for the 5-s data. As described above, the HCL output was adjusted to be near saturation a t an integration time of 5 s and was unchanged when the integration time was reduced to 1 s. Thus, the PDA output signal for 1 s should be a factor of 4.636 (the ratio of the integration times) less than for 5 s. Since the measurement precision is essentially invariant with integration time, as noted above, the relative precision decreases as the integration time decreases. However, it must be noted that it is not the integration time that is important here in regard to precision but rather how near the HCL output is to the saturation limit of the PDA a t a given integration time. If the HCL output was increased to near the PDA saturation level a t 1 8 , the measurement precision would be comparable to that obtained a t 5 s. Likewise, the differences in relative standard deviations between elements for the PDA in Table V is primarily due to the different value of E , for each element. The relative standard deviations with the P M T are always less than with the PDA because of the dominance of electronic noise with the PDA. Note that the values of i, with the PDA detector are within a factor of 2.5of those measured for a PMT system and a factor of 4 higher for Zn. Apparently the small size of the detector is compensated by summing the signals from five diodes and by using a monochromator with a lower

ANALYTICAL FIGURES OF MERIT The data obtained for the precision calculations may also be used to establish analytical figures of merit for sensitivity and detection limits. The results are presented in Table VI for the elements tested here. The flame stoichiometry and burner position were optimized for each element and the HCLs were operated at, or near, the recommended currents. Literature values shown were obtained with a burner/nebulizer assembly indentical with that employed here (17). The results of Table VI indicate that the experimental sensitivity is essentially constant as the integration time is decreased with all other instrumental parameters constant. This is to be expected since, for a given instrumental configuration, the sensitivity is a function of the analyte species, the optics, and the burner nebulization-atomization efficiency. However, the detection limit increases as the integration times is reduced. This is also to be expected since the detection limit is defined in terms of the measurement precision. As the integration time is decreased, E , decreases which worsens the precision or increases the relative noise, in the blank signal. The experimental and literature values for sensitivity compare quite favorably. Results for Cu and Mg are identical with the literature value while the remaining values are a factor of 2-3 greater than the literature values. The experimental results for the detection limits are approximately 4-10 times greater then the literature values. This is due to the reduction in relative measurement precision resulting from the electronic readout noise as discussed above. A comparison with results reported by Chaung (2) shows comparable sensitivities for Cr and Mn, however, the results obtained here for Cu, Mg, and Zn are a factor of 2-5 lower than reported by Chuang. This is undoubtedly a result of the fact that, in order to use short integration times, Chuang employed very high HCL currents of 20,20, and 12 mA for Cu, Mg, and Zn, respectively. The deleterious effects on analytical performance of such high lamp currents are well recognized. The sensitivity data reported here were obtained utilizing recommended lamp currents and hence should be

ANALYTICAL CHEMISTRY, VOL. 52, NO. 13, NOVEMBER 1980

Table VII. Figures of Merit for Simultaneous Multielement Analyses of Ag/Cu and Mg/Mn HCL current, integration sensitivity, detection ppm limit, ppm element mA time, s Ag/Cu Determination Ag CU

9.0 5.5

0.081

1

0.06 1 0.067 0.061

5

0.059

0.022

1

5

0.023

0.097

Mg/Mn Determination Mg

Mn

3.2

19.2

1

0.005

0.006

5

0.005 0.041 0.038

0.001

1 5

0.046

0.011

more indicative of the PDA-AA measurement capabilities. T h e detection limits reported by Chuang are generally comparable to those obtained here. This should be expected since identical detector and readout systems (RL512C/17 and RC100/102) were used in both studies, and hence both PDA-AA systems should be electronic readout noise limited, although Chuang did not explicitly report that to be the case. The detection limit for Mg is considerably improved because of the better sensitivity achieved with lower HCL current.

MULTIELEMENT ANALYSIS In an earlier report Horlick ( 1 )demonstrated the utility of a PDA-AA spectrometer for performing simultaneous multielement analysis and similar results for simultaneous determination of two and three elements have been reported by Chuang (2). The limiting factors on the number of elements which may be determined simultaneously are the number of HCLs which can be multiplexed, the availability of suitable multielement HCLs, and the wavelength range which can be simultaneously monitored. Using a beamsplitter to combine the outputs of lamps, as depicted in Figure 2, allows only two or three lamps to be combined due to the severe loss of throughput a t each beamsplitter. A more satisfactory method of multiplexing the HCL outputs must be devised to make this approach more attractive. T h e limitation in the wavelength range which may be simultaneously monitored is a consequence of the reciprocal linear dispersion of the monochromator and the length of the PDA being used. The combination of monochromator/PDA employed here allows a n approximately 28-nm region to be monitored. Multielement determinations were performed with Mn and Mg as analytes and with Cu and Ag as analytes. In both cases single-element lamps were used and multiplexed as shown in Figure 2 . As compensation for the loss in radiant throughput due to inserting the beamsplitter in the optical path, the lamps were operated at higher currents as noted in Table VII. The flame stoichiometry and burner positon were optimized for each pair of analytes. It was found that the analyte pair Mg/Mn exhibited maximum absorbance under identical conditions, as did the analyte pair Cu/Ag. The data acquisition and reduction procedures were as discussed above. The analytical figures of merit are presented in Table VII. A comparison of the analytical figures of merit obtained from the multielement determinations with those obtained from the single element determinations shows that they are essentially identical and hence there would be no decrease in analytical performance in the simultaneous determination of these analyte pairs. However, in the case where the analytes required markedly different analytical conditions, such as in the simultaneous determination of manganese and molybdenum ( 1 8 ) ,the analytical performance would certainly be degraded. The analytical conditions could be optimized for

2139

the determination of one of the analytes and thereby sacrifice the figures of merit for the other analyte. An alternative approach would be to establish the “best” compromised analytical conditions for the determination of both analytes simultaneously and, hence, realize somewhat inferior figures of merit for both.

IMPROVEMENTS Equation A3 can be used to estimate the characteristics of a PDA-AA spectrometer required to yield precision equivalent to a conventional AA with a P M T detector. From the electronic readout noise limiting form of eq A3, eq A12, aOt’/Er or the readout noise relative to saturation level should be 2 X or less to make readout noise nonlimiting for most elements for absorbances from 0 to 1. If the HCL signals from all diodes with a signal level greater or equal to 0.7 of the peak diode signal are summed, then aot’/E, can be larger. For five diodes as used in this work, the relative readout noise can be and still be nonlimiting. For the as great as about 4 x above statements it is assumed that the reference signal is near the saturation level for the peak diode. If it is assumed that reference voltages can be adjusted to within a factor of 5 of saturation, then the relative readout noise would have to be one-fifth of the above values to ensure no degradation of precision by readout noise for the elements with smaller HCL signals. As indicated previously, PDA readout systems have been designed with relative readout resolution for one diode of 1 X lo4 or less (14). Thus with present technology, it is possible to design a PDA-AA spectrometer in which the detector noise characteristics are nonlimiting as they are with conventional AA spectrometers with a PMT. For many elements, the relative noise from other sources (lamp, flame transmission, or analyte flicker noise) is much larger than the minimum values assumed above so that the aa’/E, could be larger but still not limiting. Also n integrations can be averaged to effectively reduce aB‘/E, by n1I2.Here, however, speed of analysis is sacrificied, and with long total integration times the relative contribution of drift and flicker noise arising from the lamp and flame could be accentuated. T h e smaller effective reference HCL signal (i,) obtained with a PDA due to the smaller size of a pixel will not cause any real problems. If the HCL signal is near saturation, the relative signal shot noise (a,)q+s/Er, as calculated from eq A13, is about 2 x and smaller than the crfi’/E, observed with a conventional AA since urt’is usually limited by lamp or flame transmission flicker noise rather than signal shot noise (8-10). With saturation, u A / A from eq A14, the shot noise limiting form of eq A3, is about 3 X at A = 1. Since aA/A a t A = 1 is typically 2-10 x due to analyte absorption flicker noise or analyte or background emission noise (&IO), signal shot noise is not limiting. Even if the HCL signal is one-fifth of saturation, the above relative standard deviations would be only 5l/* larger and usually not limiting. If t,he signals from five diodes are averaged as in this work, the total effective i, is increased by about a factor of 4 and the above RSD due to lamp shot noise is reduced by a factor of 2. Even if a diode is used below saturation it is more likely, even in a situation with lower electronic readout noise, that readout noise rather than lamp shot noise will be limiting since the absolute signal shot noise drops in proportion to the square root of the signal whereas the electronic noise is signal independent. For multielement work, the above noise considerations apply if all used HCL diode signals are within a factor of 5 of the largest used signal. This depends upon a fortuitous choice of elements in one multielement lamp or combining the beams from several lamps. The later approach is probably limited to four elements (two or less beam splitters per HCL because i, will be as low as 20% (13) of what it is for an

2140

ANALYTICAL CHEMISTRY, VOL. 52, NO. 13, NOVEMBER 1980

individual HCL and thus increase the relative readout noise by up to a factor of 5. This can be compensated for by longer integration times or higher lamp currents which both have negative aspects as previously indicated. The use of the bigger 2.5-mm PDA in place of the 0.43-mm high PDA used in this work will increase the reference signal (i,) by over a factor of 5 if all other conditions are equivalent. This means that equivalent precision could be obtained in one-fifth of the time or that in the same time, by averaging five integration times, precision can be improved by over a factor of 2. Thus with low relative readout noise electronics, the larger PDA should provide equivalent performance to a conventional AA in the same time except where analyte or background emission signal or noise are limiting. The larger array also makes multielement analysis more feasible since saturation could be observed in reasonable integration times without significantly increasing HCL currents.

ir

T K “e

n t1

E2

“Ot F3 “Ot‘

“be

e

t G G‘ n a (ar)q+s

lamp photocathodic current, A transmittance bandwidth constant, A root-mean-square noise in analyte emission signal, V gain of photomultiplier, dimensionless source flicker factor, dimensionless flame transmission flicker factor, dimensionless 0% T root-mean-square noise, V analyte absorption flicker factor, dimensionless 0% T‘ root-mean-square noise, V root-mean-square noise in background emission signal, V 1.6 x 10-19 c integration time, s amplification factor for P M T system, V/A amplification factor for diode array system, V/C number of integrations secondary emission (relative noise due to the statistical fluctuation in P M T gain), dimensionless lamp shot noise, V

LITERATURE CITED

got2

= (goo2

+ (5be2

K = e(1 + a ) / T E , = i,mG Specifically for the PDA: G = G’T

m = l a=o K = e/tn E, = i,G’t

+ T2

aA/A = (ao4/Er)(2(1

-

T1))’I2/(-ln T )

( u ~ ) ~ +=~( e/ /E t i r~) ’ i 2 uA/A = ( e / t i r ) ” ’ ( ( 1 + T 1 ) / ( - l n r ) ) A UA

GLOSSARY absorbance standard deviation in absorbance

Horlick, G.; Codding, E. G. Appl. Spectrosc. 1975, 29, 167-170. Chuang, F. S.; Natusch, D. F. S.; O’Keefe, K.R. Anal. Chem. 1976, 50, 525-530. Jackson, K. W.; Aldous, K. M.; Mitchell, D. G. Spectrosc. Lett. 1973, 6 , 315-321. Mitchell, D. G.;Jackson, K. W.; Aldous, K. M. Anal. Chem. 1973, 45. 1215A-1223A. Jackson, K. W.; Aldous, K. M.; Mitchell. D. G. Appl. Spectrosc. 1974, 28 569-573. ... Aldous, K. M ; Mitchell, D. G.: Jackson, K. W. Anal. Chem. 1975, 47, 1034-1 037. Ingle, J. D., Jr. Anal. Chem. 1974, 4 6 , 2161-2171. Bower, N. W.; Ingle, J. D., Jr. Anal. Chem. 1976, 48, 686-692. Bower, N. W.; Ingle, J. D., Jr. Anal. Chem. 1977, 49, 574-579. Bower, N. W.; Ingle, J. D., Jr. Anal. Chem. 1979, 51, 72-76. Horlick. G. Appl. Spectrosc. 1976, 30, 113-123. Horlick. G.;Codding, E. G. I n “Contemporary Topics In Analytical and Clinical Chemistry”; Hercules, D. H., Hieftje, G. M., Snyder, L. R., Evenson, M. A., Eds.; Plenum: New York, 1977; Vol. 1, Chapter 5. Salin, E. D.; Ingle, J. D., Jr. Anal. Chem. 1976, 5 0 , 1745-1750. Simpson, R. W. Rev. Scl. Instrum. 1979, 50. 730-732. Talmi, Yair; Simpson, R. W. Appl. Opt. 1980, 19, 1401-1414. Bower, N. W.; Ingle, J. D., Jr. Anal. Chim. Acta 1979, 105, 199-212. Parker, C. R. ”Water Analysis by Atomic Absorption Spectroscopy”; Varian Techtron: Palo Alto, CA, 1972. Brost, D. F.; Mallory, D.; Busch, K. W. Anal. Chem. 1977, 49, 2280-2284.

-..

~

RECEIVED for review January 3,1980. Accepted July 28,1980. Financial support by the University of Calgary and the Natural Sciences and Engineering Research Council of Canada is gratefully acknowledged. This work was presented in part as paper no. AN-27 a t the 62nd Canadian Chemical Conference, Vancouver, British Columbia, 1979.